Pyromania: Fire transforms terrestrial subsidies in aquatic ecosystems
CB Wall
5/5/2023
Project overview
The goal of the Pyromania project is to test how terrestrial subsides (plant biomass loading or “browning”) and burning influence aquatic productivity, water quality/chemistry, and trophic transfer. We used a manipulative experiment to assess a range of plant material quantities (0-400g per tank) and fire treatment (burned vs unburned material) and the non-linearity of these effects on aquatic systems through 4 time-point samplings. We used 400L aquatic mesocosms and ran the experiment for ~90d in 2021-2022.
Figure 1. Pyromania experimental setup
DATA SETS
This data set is among 3 to be generated for the project and focuses
on:
- dissolved organic carbon (DOC) concentrations
- total dissolved nitrogen (TDN) concentrations
- net ecosystem productivity and respiration (NEP, R)
using whole mesocosm dissolved oxygen (O2) measurements
- burning effects on donor plant material using
elemental analysis
- greenhouse gas concentrations (carbon dioxide
(CO2), methane (CH4)) from mesocosms
- trophic transfer using 15N-labeled sage plants to trace N from the plants into plankton.
TIME POINTS
Time-0 (T0) = before the addition of plant
materials, plankton were stocked at this stage
Time-1 (T1) = 10 days after the addition of plant
materials
Time-2 (T2) = 31 days after …
Time-3 (T3) = 59 days after …
Time-4 (T4) = 89 days after …
General notes on GAM analyses
We fit the generalized additive models (GAMs) via
restricted maximum likelihood (REML) to give stable results
with the smoothing parameter (sp) to determine the
non-linear relationship between response variables and plant-biomass
loading (x-axis). We use automatic smoothing with k value
generated automatically from the models, which will set the line
‘wiggliness’. Too low and the relationship becomes linear; too high, and
the wiggliness goes haywire.
When using the non-linear smoothing, this is the s(x).
When the variable is inside the smooth function, this accounts for the
nonlinear shape. We do not use additive non-linear smoothing,
which is when two smoothers together, as s(x1) + s(x2),
instead we use factor-smooth interaction (detailed below). In
addition, we use Treatment (and occasionally plankton size fractions, or
Type) as predictors outside of the smooth terms
s(x1); this allows for linearity. Continuous variables are
rarely linear in GAMs, however, setting categorical variables as linear
predictors is more common.
Factor-smooth interactions are written as
s(x1 by = fac). This sets different smoothers for different
variables of “fac”. Usually, the different smoothers are combined with a
varying intercept in case the different categories are different in
means and slopes, this would be by adding the
fac + s(x1 by = fac), where the +fac allows
for a different slope. Similarly, in the absence of
by = fac, the smoother is considered a global smoother
s(x1), fitting a single line to all the data. If a global
smoother is combined with a factor term, then this is akin to varying
the intercept but keeping the same slope: fac + s(x1).
The EDF - effective degrees of freedom equate with
wiggliness, where edf =1 is a straight line, and higher edfs as more
wiggly. GAM smoother significance is described as not being able to draw
a horizontal line through the data. Finally, it is also advised to check
model concurvity, which is the collinearity with models from 0-1.
DOC
Import the data for DOC and TDN and do a loop to clean up all files and make stacked data in single df. This will take the raw data files, align metadata, and filter to make a new df for models.
detach("package:dplyr", unload = TRUE)
library(dplyr)
## import treatment IDs
IDs<-read.csv("data/treatment.IDs.csv")
##### grab files in a list
Total.DOC.files <- list.files(path="data/DOC.TN", pattern = "csv$", full.names = T)
##### what are the file names, sans extensions using package 'tools'
file.names<-file_path_sans_ext(list.files(path="data/DOC.TN", pattern = "csv$", full.names = F))
############ formatting all data in for loop
for(i in 1:length(Total.DOC.files))
{
data<-read.csv(Total.DOC.files[i], sep=",")
data<-data[,c(1,3,4)] # removed columns we don't need
data$File<-Total.DOC.files[i]
colnames(data)<-c("Tank", "DOC..mg.L", "TN..mg.L", "File")
data$Tank<- IDs$Tank
data$Tank<-as.numeric(as.character(data$Tank)) # make the column of samples all numeric
data <- data[!is.na(as.numeric(as.character(data$Tank))),] # remove all rows that aren't numeric/tanks
data$Treatment<-IDs$Treatment
data$plant.mass..g<-IDs$plant.mass..g
make.names(assign(paste(file.names[i], sep=""), data)) # make the file name the name of new df for loop df
}
########## this is the end of the loop
#see all dfs you've made, the above will be df matching their file names
# ls()
#Combine files from loop to single df
DOC.df<-rbind(DOC_T0, DOC_T1, DOC_T2, DOC_T3, DOC_T4)
DOC.df$File <- sapply(strsplit(DOC.df$File, "/"), `[`, 3) # extract sample names
# alternative way to code the above
#give the 10th-24th character of the file name, removing the rest
#DOC.df$File<-substr(DOC.df$File, 13, 27)
#alternatively
# remove the 9 letters ('^.) at start
# remove the 4 letters (.$') at end
#DOC.df$File<-gsub('^.........|....$', '', DOC.df$File)
# if equals DOC_T0_11052021 then, T0, if not then T1
DOC.df$Time.point<- as.factor(ifelse(DOC.df$File=="DOC_T0.csv", "T0",
ifelse (DOC.df$File=="DOC_T1.csv", "T1",
ifelse (DOC.df$File=="DOC_T2.csv", "T2",
ifelse(DOC.df$File=="DOC_T3.csv", "T3", "T4")))))
#rearrange
DOC.df<- DOC.df %>%
select(File, Time.point, Treatment, Tank, plant.mass..g, DOC..mg.L, TN..mg.L)
DOC.df$Treatment<-as.factor(DOC.df$Treatment)Analyze DOC at each time point. Run model selection and produce plots for each individual timepoint, later pooled into a 5 panel figure.
######## T0 model
m1.DOC.T0<-gam(DOC..mg.L ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T0", data = DOC.df, method = "REML")
m2.DOC.T0<-gam(DOC..mg.L ~ Treatment + s(plant.mass..g), subset = Time.point=="T0", data = DOC.df, method = "REML")
m3.DOC.T0<-gam(DOC..mg.L ~ s(plant.mass..g), subset = Time.point=="T0", data = DOC.df, method = "REML")
T0.DOC.AIC<-AIC(m1.DOC.T0, m2.DOC.T0, m3.DOC.T0)
# best is smoother solo
summary(m3.DOC.T0)
anova.gam(m3.DOC.T0)
gam.check(m3.DOC.T0, rep=1000)
draw(m3.DOC.T0)
concrvity(m3.DOC.T0)
par(mfrow = c(2, 2))
plot(m3.DOC.T0, all.terms = TRUE, page=1)
# model predictions
DOC.diff.T0<-plot_difference(
m1.DOC.T0,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
DOC.T0.mod.plot<-
plot_smooths(
model = m3.DOC.T0,
series = plant.mass..g,
) +
geom_point(data=DOC.df[(DOC.df$Time.point=="T0"),],
aes(x=plant.mass..g, y=DOC..mg.L, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
coord_cartesian(ylim=c(0, 60)) +
ggtitle("Day-0") +
ylab("DOC (mg/L)") +
xlab("plant material (g)") +
Fig.formatting
######## T1 model
m1.DOC.T1<-gam(DOC..mg.L ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T1", data = DOC.df, method = "REML")
m2.DOC.T1<-gam(DOC..mg.L ~ Treatment + s(plant.mass..g), subset = Time.point=="T1", data = DOC.df, method = "REML")
m3.DOC.T1<-gam(DOC..mg.L ~ s(plant.mass..g), subset = Time.point=="T1", data = DOC.df, method = "REML")
T1.DOC.AIC<-AIC(m1.DOC.T1, m2.DOC.T1, m3.DOC.T1)
# best is smooth by factor
summary(m1.DOC.T1)
anova.gam(m1.DOC.T1)
gam.check(m1.DOC.T1, rep=1000)
draw(m1.DOC.T1)
concrvity(m1.DOC.T1)
par(mfrow = c(2, 2))
plot(m1.DOC.T1, all.terms = TRUE, page=1)
# model predictions
DOC.diff.T1<-plot_difference(
m1.DOC.T1,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
DOC.T1.mod.plot<-
plot_smooths(
model = m1.DOC.T1,
series = plant.mass..g,
comparison = Treatment
) +
geom_point(data=DOC.df[(DOC.df$Time.point=="T1"),],
aes(x=plant.mass..g, y=DOC..mg.L, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
coord_cartesian(ylim=c(0, 60)) +
ggtitle("Day-10") +
ylab("DOC (mg/L)") +
xlab("plant material (g)") +
Fig.formatting
# effect of treatment, smoothing significant across both treatments
# DOC higher in unburned, relative to burned
########## T2
m1.DOC.T2<-gam(DOC..mg.L ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T2", data = DOC.df, method = "REML")
m2.DOC.T2<-gam(DOC..mg.L ~ Treatment + s(plant.mass..g), subset = Time.point=="T2", data = DOC.df, method = "REML")
m3.DOC.T2<-gam(DOC..mg.L ~ s(plant.mass..g), subset = Time.point=="T2", data = DOC.df, method = "REML")
T2.DOC.AIC<-AIC(m1.DOC.T2, m2.DOC.T2, m3.DOC.T2)
# best is smooth by factor
summary(m1.DOC.T2)
anova.gam(m1.DOC.T2)
gam.check(m1.DOC.T2, rep=1000)
draw(m1.DOC.T2)
concrvity(m1.DOC.T2)
par(mfrow = c(2, 2))
plot(m1.DOC.T2, all.terms = TRUE, page=1)
# model predictions
DOC.diff.T2<-plot_difference(
m1.DOC.T2,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
DOC.T2.mod.plot<-
plot_smooths(
model = m1.DOC.T2,
series = plant.mass..g,
comparison = Treatment
) +
geom_point(data=DOC.df[(DOC.df$Time.point=="T2"),],
aes(x=plant.mass..g, y=DOC..mg.L, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
coord_cartesian(ylim=c(0, 60)) +
ggtitle("Day-31") +
ylab("DOC (mg/L)") +
xlab("plant material (g)") +
Fig.formatting
# NO effect of treatment, smoothing significant across both treatments
# DOC equivalent in burned and unburned
# DOC more variable/wonky across gradient in burned
########## T3
m1.DOC.T3<-gam(DOC..mg.L ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T3", data = DOC.df, method = "REML")
m2.DOC.T3<-gam(DOC..mg.L ~ Treatment + s(plant.mass..g), subset = Time.point=="T3", data = DOC.df, method = "REML")
m3.DOC.T3<-gam(DOC..mg.L ~ s(plant.mass..g), subset = Time.point=="T3", data = DOC.df, method = "REML")
T3.DOC.AIC<-AIC(m1.DOC.T3, m2.DOC.T3, m3.DOC.T3)
# best by factor smooth
summary(m1.DOC.T3)
anova.gam(m1.DOC.T3)
gam.check(m1.DOC.T3, rep=1000)
draw(m1.DOC.T3)
concrvity(m1.DOC.T3)
par(mfrow = c(2, 2))
plot(m1.DOC.T3, all.terms = TRUE, page=1)
# model predictions
DOC.diff.T3<-plot_difference(
m1.DOC.T3,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
DOC.T3.mod.plot<-
plot_smooths(
model = m1.DOC.T3,
series = plant.mass..g,
comparison = Treatment
) +
geom_point(data=DOC.df[(DOC.df$Time.point=="T3"),],
aes(x=plant.mass..g, y=DOC..mg.L, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
coord_cartesian(ylim=c(0, 60)) +
ggtitle("Day-59") +
ylab("DOC (mg/L)") +
xlab("plant material (g)") +
Fig.formatting
# effect of treatment, smoothing significant across both treatments
# DOC higher in burned vs. unburned
########## T4
m1.DOC.T4<-gam(DOC..mg.L ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T4", data = DOC.df, method = "REML")
m2.DOC.T4<-gam(DOC..mg.L ~ Treatment + s(plant.mass..g), subset = Time.point=="T4", data = DOC.df, method = "REML")
m3.DOC.T4<-gam(DOC..mg.L ~ s(plant.mass..g), subset = Time.point=="T4", data = DOC.df, method = "REML")
T4.DOC.AIC<-AIC(m1.DOC.T4, m2.DOC.T4, m3.DOC.T4)
# best is global
summary(m3.DOC.T4)
anova.gam(m3.DOC.T4)
gam.check(m3.DOC.T4, rep=1000)
draw(m3.DOC.T4)
concrvity(m3.DOC.T4)
par(mfrow = c(2, 2))
plot(m3.DOC.T4, all.terms = TRUE, page=1)
# model predictions
DOC.diff.T4<-plot_difference(
m1.DOC.T4,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
DOC.T4.mod.plot<-
plot_smooths(
model = m3.DOC.T4,
series = plant.mass..g
) +
geom_point(data=DOC.df[(DOC.df$Time.point=="T4"),],
aes(x=plant.mass..g, y=DOC..mg.L, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
coord_cartesian(ylim=c(0, 60)) +
ggtitle("Day-89") +
ylab("DOC (mg/L)") +
xlab("plant material (g)") +
Fig.formatting
# no effect of treatment, smoothing significant across both treatments
# DOC equivalent in burned and unburned
mod.rep<-rep(c("~Treatment + s(plant.mass..g, by= Treatment)",
"~Treatment + s(plant.mass..g)",
"~s(plant.mass..g)"), times=5)
mod.DOC.df<- data.frame(mod.rep)
AIC.DOC<-bind_rows(T0.DOC.AIC, T1.DOC.AIC, T2.DOC.AIC, T3.DOC.AIC, T4.DOC.AIC)
AIC.DOC.mod<-cbind(mod.DOC.df, AIC.DOC)
write.csv(AIC.DOC.mod, "output/AIC models/AIC.DOC.csv")DOC tables
Table: Results for DOC Time-0
anova.gam(m3.DOC.T0)##
## Family: gaussian
## Link function: identity
##
## Formula:
## DOC..mg.L ~ s(plant.mass..g)
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g) 1.04 1.08 1.341 0.24Table: Results for DOC Time-1
anova.gam(m1.DOC.T2)##
## Family: gaussian
## Link function: identity
##
## Formula:
## DOC..mg.L ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric Terms:
## df F p-value
## Treatment 1 0.035 0.853
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 6.482 7.532 34.39 <2e-16
## s(plant.mass..g):Treatmentunburned 1.568 1.929 59.34 <2e-16Table: Results for DOC Time-2
anova.gam(m1.DOC.T2)##
## Family: gaussian
## Link function: identity
##
## Formula:
## DOC..mg.L ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric Terms:
## df F p-value
## Treatment 1 0.035 0.853
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 6.482 7.532 34.39 <2e-16
## s(plant.mass..g):Treatmentunburned 1.568 1.929 59.34 <2e-16Table: Results for DOC Time-3
anova.gam(m1.DOC.T3)##
## Family: gaussian
## Link function: identity
##
## Formula:
## DOC..mg.L ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric Terms:
## df F p-value
## Treatment 1 12.32 0.00182
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 2.051 2.532 94.00 <2e-16
## s(plant.mass..g):Treatmentunburned 2.202 2.714 56.55 <2e-16Table: Results for DOC Time-4
anova.gam(m3.DOC.T4)##
## Family: gaussian
## Link function: identity
##
## Formula:
## DOC..mg.L ~ s(plant.mass..g)
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g) 1.928 2.385 29.8 <2e-16DOC figures
Compile raw plots and model-diff plots for final figures.
###### compile the plots with effect plots
extract.legend <- get_legend(
# create some space to the left of the legend
DOC.T1.mod.plot + theme(legend.box.margin = margin(0, 0, 0, 10)))
DOC.alltimes<-plot_grid(
DOC.T0.mod.plot+ theme(legend.position = "none"),
DOC.T1.mod.plot+ theme(legend.position = "none"),
DOC.T2.mod.plot+ theme(legend.position = "none"),
DOC.T3.mod.plot+ theme(legend.position = "none"),
DOC.T4.mod.plot+ theme(legend.position = "none"), extract.legend,
rel_widths = c(8,8,8,8,8,3), ncol=6)
ggsave("figures/DOC.alltimes.mod.pdf", height=4, width=15)
DOC.alltimes
Model differences between the two factor-smoothers. The areas in pink show where there are significant differences between the two smoothers, indicating treatment effects.
DOC.mod.diffs<-plot_grid(
DOC.diff.T0+ theme(legend.position = "none")+ ggtitle("DOC - Day-0"),
DOC.diff.T1+ theme(legend.position = "none")+ ggtitle("DOC - Day-10"),
DOC.diff.T2+ theme(legend.position = "none")+ ggtitle("Day-31"),
DOC.diff.T3+ theme(legend.position = "none")+ ggtitle("Day-59"),
DOC.diff.T4+ theme(legend.position = "none")+ ggtitle("Day-89"),
rel_widths = c(8,8,8,8,8), ncol=5)
ggsave("figures/DOC.mod.diffs.pdf", height=4, width=14)
DOC.mod.diffs
TDN
Total N analysis and plots, running models and making model-diff plots.
TN.df<-DOC.df
######## T0 model
m1.TN.T0<-gam(TN..mg.L ~ Treatment + s(plant.mass..g, by=Treatment),
subset = Time.point=="T0", data = TN.df, method = "REML")
m2.TN.T0<-gam(TN..mg.L ~ Treatment + s(plant.mass..g),
subset = Time.point=="T0", data = TN.df, method = "REML")
m3.TN.T0<-gam(TN..mg.L ~ s(plant.mass..g),
subset = Time.point=="T0", data = TN.df, method = "REML")
T0.TN.AIC<-AIC(m1.TN.T0, m2.TN.T0, m3.TN.T0)
# best smooth by factor
summary(m1.TN.T0)
anova.gam(m1.TN.T0)
gam.check(m1.TN.T0, rep=1000)
draw(m1.TN.T0)
concrvity(m1.TN.T0)
par(mfrow = c(2, 2))
plot(m1.TN.T0, all.terms = TRUE, page=1)
# model predictions
TN.diff.T0<-plot_difference(
m1.TN.T0,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
## plot for the model output on rawdata
TN.T0.mod.plot<-
plot_smooths(
model = m1.TN.T0,
series = plant.mass..g,
comparison = Treatment
) +
geom_point(data=TN.df[(TN.df$Time.point=="T0"),],
aes(x=plant.mass..g, y=TN..mg.L, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
coord_cartesian(ylim=c(0, 2)) +
ggtitle("Day-0") +
ylab("TN (mg/L)") +
xlab("plant material (g)") +
Fig.formatting
# no treatment effect, compare to simplified model (p=0.901)
######## T1 model
m1.TN.T1<-gam(TN..mg.L ~ Treatment + s(plant.mass..g, by=Treatment),
subset = Time.point=="T1", data = TN.df, method = "REML")
m2.TN.T1<-gam(TN..mg.L ~ Treatment + s(plant.mass..g),
subset = Time.point=="T1", data = TN.df, method = "REML")
m3.TN.T1<-gam(TN..mg.L ~ s(plant.mass..g),
subset = Time.point=="T1", data = TN.df, method = "REML")
T1.TN.AIC<-AIC(m1.TN.T1, m2.TN.T1, m3.TN.T1)
#best global only
summary(m3.TN.T1)
anova.gam(m3.TN.T1)
gam.check(m3.TN.T1, rep=1000)
draw(m3.TN.T1)
concrvity(m3.TN.T1)
par(mfrow = c(2, 2))
plot(m3.TN.T1, all.terms = TRUE, page=1)
# model predictions
TN.diff.T1<-plot_difference(
m1.TN.T1,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
## plot for the model output on rawdata
TN.T1.mod.plot<-
plot_smooths(
model = m3.TN.T1,
series = plant.mass..g
) +
geom_point(data=TN.df[(TN.df$Time.point=="T1"),],
aes(x=plant.mass..g, y=TN..mg.L, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
coord_cartesian(ylim=c(0, 2)) +
ggtitle("Day-10") +
ylab("TN (mg/L)") +
xlab("plant material (g)") +
Fig.formatting
# TN smoother significant for unburned but not burned (p=0.007)
######## T2 model
m1.TN.T2<-gam(TN..mg.L ~ Treatment + s(plant.mass..g, by=Treatment),
subset = Time.point=="T2", data = TN.df, method = "REML")
m2.TN.T2<-gam(TN..mg.L ~ Treatment + s(plant.mass..g),
subset = Time.point=="T2", data = TN.df, method = "REML")
m3.TN.T2<-gam(TN..mg.L ~ s(plant.mass..g),
subset = Time.point=="T2", data = TN.df, method = "REML")
T2.TN.AIC<-AIC(m1.TN.T2, m2.TN.T2, m3.TN.T2)
#best global with treatment term
summary(m2.TN.T2)
anova.gam(m2.TN.T2)
gam.check(m2.TN.T2, rep=1000)
draw(m2.TN.T2)
concrvity(m2.TN.T2)
par(mfrow = c(2, 2))
plot(m2.TN.T2, all.terms = TRUE, page=1)
# model predictions
TN.diff.T2<-plot_difference(
m1.TN.T2,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
## plot for the model output on rawdata
TN.T2.mod.plot<-
plot_smooths(
model = m2.TN.T2,
series = plant.mass..g,
comparison=Treatment
) +
geom_point(data=TN.df[(TN.df$Time.point=="T2"),],
aes(x=plant.mass..g, y=TN..mg.L, color=Treatment)) +
geom_line(aes(fill=Treatment, linetype=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
coord_cartesian(ylim=c(0, 2)) +
ggtitle("Day-31") +
ylab("TN (mg/L)") +
xlab("plant material (g)") +
Fig.formatting
# Near treatment effect p=0.053, higher TN in unburned
# smoother signif: 0.042
######## T3 model
m1.TN.T3<-gam(TN..mg.L ~ Treatment + s(plant.mass..g, by=Treatment),
subset = Time.point=="T3", data = TN.df, method = "REML")
m2.TN.T3<-gam(TN..mg.L ~ Treatment + s(plant.mass..g),
subset = Time.point=="T3", data = TN.df, method = "REML")
m3.TN.T3<-gam(TN..mg.L ~ s(plant.mass..g),
subset = Time.point=="T3", data = TN.df, method = "REML")
T3.TN.AIC<-AIC(m1.TN.T3, m2.TN.T3, m3.TN.T3)
#best with smooth by factor term
summary(m1.TN.T3)
anova.gam(m1.TN.T3)
gam.check(m1.TN.T3, rep=1000)
draw(m1.TN.T3)
concrvity(m1.TN.T3)
par(mfrow = c(2, 2))
plot(m1.TN.T3, all.terms = TRUE, page=1)
# model predictions
TN.diff.T3<-plot_difference(
m1.TN.T3,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
## plot for the model output on rawdata
TN.T3.mod.plot<-
plot_smooths(
model = m1.TN.T3,
series = plant.mass..g,
comparison = Treatment
) +
geom_point(data=TN.df[(TN.df$Time.point=="T3"),],
aes(x=plant.mass..g, y=TN..mg.L, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
coord_cartesian(ylim=c(0, 2)) +
ggtitle("Day-59") +
ylab("TN (mg/L)") +
xlab("plant material (g)") +
Fig.formatting
# Near treatment effect p=0.075, trend for higher TN in burned
# smoother signif: at <0.001 for both
######## T4 model
m1.TN.T4<-gam(TN..mg.L ~ Treatment + s(plant.mass..g, by=Treatment),
subset = Time.point=="T4", data = TN.df, method = "REML")
m2.TN.T4<-gam(TN..mg.L ~ Treatment + s(plant.mass..g),
subset = Time.point=="T4", data = TN.df, method = "REML")
m3.TN.T4<-gam(TN..mg.L ~ s(plant.mass..g),
subset = Time.point=="T4", data = TN.df, method = "REML")
T4.TN.AIC<-AIC(m1.TN.T4, m2.TN.T4, m3.TN.T4)
#best with smooth by factor term
summary(m1.TN.T4)
anova.gam(m1.TN.T4)
gam.check(m1.TN.T4, rep=1000)
draw(m1.TN.T4)
concrvity(m1.TN.T4)
par(mfrow = c(2, 2))
plot(m1.TN.T4, all.terms = TRUE, page=1)
# model predictions
TN.diff.T4<-plot_difference(
m1.TN.T4,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
## plot for the model output on rawdata
TN.T4.mod.plot<-
plot_smooths(
model = m1.TN.T4,
series = plant.mass..g,
comparison = Treatment
) +
geom_point(data=TN.df[(TN.df$Time.point=="T4"),],
aes(x=plant.mass..g, y=TN..mg.L, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
coord_cartesian(ylim=c(0, 2)) +
ggtitle("Day-89") +
ylab("TN (mg/L)") +
xlab("plant material (g)") +
Fig.formatting
# effect of treatment (higher TN in the burned) (p=0.020)
# significant smoother effect for burned treatment only (p=0.032)
mod.rep<-rep(c("~Treatment + s(plant.mass..g, by= Treatment)",
"~Treatment + s(plant.mass..g)",
"~s(plant.mass..g)"), times=5)
mod.TN.df<- data.frame(mod.rep)
AIC.TN<-bind_rows(T0.TN.AIC, T1.TN.AIC, T2.TN.AIC, T3.TN.AIC, T4.TN.AIC)
AIC.TN.mod<-cbind(mod.TN.df, AIC.TN)
write.csv(AIC.TN.mod, "output/AIC models/AIC.TN.mod.csv")
TDN tables
Results for TDN Time-0
anova.gam(m1.TN.T0)##
## Family: gaussian
## Link function: identity
##
## Formula:
## TN..mg.L ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric Terms:
## df F p-value
## Treatment 1 0.879 0.357
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 1 1 0.009 0.9266
## s(plant.mass..g):Treatmentunburned 1 1 6.303 0.0186Table: Results for TDN Time-1
anova.gam(m3.TN.T1)##
## Family: gaussian
## Link function: identity
##
## Formula:
## TN..mg.L ~ s(plant.mass..g)
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g) 2.848 3.492 5.72 0.00274Table: Results for TDN Time-2
anova.gam(m2.TN.T2)##
## Family: gaussian
## Link function: identity
##
## Formula:
## TN..mg.L ~ Treatment + s(plant.mass..g)
##
## Parametric Terms:
## df F p-value
## Treatment 1 4.122 0.0532
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g) 3.207 3.921 4.87 0.00425Table: Results for TDN Time-3
anova.gam(m1.TN.T3)##
## Family: gaussian
## Link function: identity
##
## Formula:
## TN..mg.L ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric Terms:
## df F p-value
## Treatment 1 3.5 0.0752
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 4.359 5.269 23.03 < 2e-16
## s(plant.mass..g):Treatmentunburned 2.457 3.022 10.52 0.000194Table: Results for TDN Time-4
anova.gam(m1.TN.T4)##
## Family: gaussian
## Link function: identity
##
## Formula:
## TN..mg.L ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric Terms:
## df F p-value
## Treatment 1 6.231 0.0196
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 2.417 2.973 3.613 0.0318
## s(plant.mass..g):Treatmentunburned 1.000 1.000 0.531 0.4731TDN figures
Compile raw plots and model-diff plots for final figures.
###### compile the plots with effect plots
TN.mod.alltimes<-plot_grid(
TN.T0.mod.plot+ theme(legend.position = "none"),
TN.T1.mod.plot+ theme(legend.position = "none"),
TN.T2.mod.plot+ theme(legend.position = "none"),
TN.T3.mod.plot+ theme(legend.position = "none"),
TN.T4.mod.plot+ theme(legend.position = "none"), extract.legend,
rel_widths = c(8,8,8,8,8,3), ncol=6)
ggsave("figures/TN.mods.plots.pdf", height=4, width=13)
TN.mod.alltimes
Model differences between the two factor-smoothers. The areas in pink show where there are significant differences between the two smoothers, indicating treatment effects.
TN.mod.diffs<-plot_grid(
TN.diff.T0+ theme(legend.position = "none")+ ggtitle("Day-0"),
TN.diff.T1+ theme(legend.position = "none")+ ggtitle("Day-10"),
TN.diff.T2+ theme(legend.position = "none")+ ggtitle("Day-31"),
TN.diff.T3+ theme(legend.position = "none")+ ggtitle("Day-59"),
TN.diff.T4+ theme(legend.position = "none")+ ggtitle("Day-89"),
rel_widths = c(8,8,8,8,8,3), ncol=6)
ggsave("figures/TN.mod.diffs.pdf", height=4, width=13)
TN.mod.diffs
YSI
Import YSI data and produce plots of changes in O2% and net ecosystem productivity (NEP) and respiration (R). The YSI data includes Temp, pH, dissolved oxygen (percent and concentration), and conductivity. Here, we will pull in the raw data and make the new metris NEP and R, determined from differences in DO% from dawn-dusk (NEP) and dusk-dawn (R) over a 24h period in each time period.
#load YSI data
YSI<-read.csv("data/Pyro_YSI.csv")
# fix date
YSI$Date<-as.character(YSI$Date)
YSI$Date<-as.POSIXct(YSI$Date, format="%m/%d/%Y")
YSI$Date<-as.Date(YSI$Date, format="%m/%d/%Y")
####### Time 1 change in O2 ################
#separate time points
YSI.T1<- YSI[(YSI$Time.point=="T1"),]
#calculate NEP for T1
T1.Prod<-YSI.T1[(YSI.T1$Date == "2021-11-15"),] # dawn and dusk for 12h period
T1.Dawn1<-T1.Prod[(T1.Prod$Dawn..Dusk == "dawn"),] # dawn-1 measurements
T1.Dusk<-T1.Prod[(T1.Prod$Dawn..Dusk == "dusk"),] # dusk measurements
T1.Dawn2<-YSI.T1[(YSI.T1$Date == "2021-11-16"),] # dawn-2 measurements, following AM
# make new dataframe
T1.O2<-(T1.Dawn1[,c(2,4:6)])
T1.O2$dawn1<-T1.Dawn1$DO.percent
T1.O2$dusk1<-T1.Dusk$DO.percent
T1.O2$dawn2<-T1.Dawn2$DO.percent
# R = dusk - dawn (PM to AM, O2 change of day 1)
# NEP = dusk - dawn (PM to AM, O2 change of day 2)
T1.O2<- mutate(T1.O2,
NEP=dusk1 - dawn1,
R=dawn2 - dusk1)
#sort
T1.O2<-T1.O2 %>%
arrange(Treatment, plant.mass..g)
################ ################ ################
####### Time 2 change in O2 ################
#separate time points
YSI.T2<- YSI[(YSI$Time.point=="T2"),]
#calculate NEP for T2
T2.Prod<-YSI.T2[(YSI.T2$Date == "2021-12-06"),] # dawn and dusk for 12h period
T2.Dawn1<-T2.Prod[(T2.Prod$Dawn..Dusk == "dawn"),] # dawn-1 measurements
T2.Dusk<-T2.Prod[(T2.Prod$Dawn..Dusk == "dusk"),] # dusk measurements
T2.Dawn2<-YSI.T2[(YSI.T2$Date == "2021-12-07"),] # dawn-2 measurements, following AM
# make new dataframe
T2.O2<-(T2.Dawn1[,c(2,4:6)])
T2.O2$dawn1<-T2.Dawn1$DO.percent
T2.O2$dusk1<-T2.Dusk$DO.percent
T2.O2$dawn2<-T2.Dawn2$DO.percent
T2.O2<- mutate(T2.O2,
NEP=dusk1 - dawn1,
R=dawn2 - dusk1)
#sort
T2.O2<-T2.O2 %>%
arrange(Treatment, plant.mass..g)
################ ################ ################
####### Time 3 change in O2 ################
#separate time points
YSI.T3<- YSI[(YSI$Time.point=="T3"),]
#calculate NEP for T3
T3.Prod<-YSI.T3[(YSI.T3$Date == "2022-01-03"),] # dawn and dusk for 12h period
T3.Dawn1<-T3.Prod[(T3.Prod$Dawn..Dusk == "dawn"),] # dawn-1 measurements
T3.Dusk<-T3.Prod[(T3.Prod$Dawn..Dusk == "dusk"),] # dusk measurements
T3.Dawn2<-YSI.T3[(YSI.T3$Date == "2022-01-04"),] # dawn-2 measurements, following AM
# make new dataframe
T3.O2<-(T3.Dawn1[,c(2,4:6)])
T3.O2$dawn1<-T3.Dawn1$DO.percent
T3.O2$dusk1<-T3.Dusk$DO.percent
T3.O2$dawn2<-T3.Dawn2$DO.percent
T3.O2<- mutate(T3.O2,
NEP=dusk1 - dawn1,
R=dawn2 - dusk1)
#sort
T3.O2<-T3.O2 %>%
arrange(Treatment, plant.mass..g)
################ ################ ################
####### Time 3 change in O2 ################
#separate time points
YSI.T4<- YSI[(YSI$Time.point=="T4"),]
#calculate NEP for T4
T4.Prod<-YSI.T4[(YSI.T4$Date == "2022-02-02"),] # dawn and dusk for 12h period
T4.Dawn1<-T4.Prod[(T4.Prod$Dawn..Dusk == "dawn"),] # dawn-1 measurements
T4.Dusk<-T4.Prod[(T4.Prod$Dawn..Dusk == "dusk"),] # dusk measurements
T4.Dawn2<-YSI.T4[(YSI.T4$Date == "2022-02-03"),] # dawn-2 measurements, following AM
# make new dataframe
T4.O2<-(T4.Dawn1[,c(2,4:6)])
T4.O2$dawn1<-T4.Dawn1$DO.percent
T4.O2$dusk1<-T4.Dusk$DO.percent
T4.O2$dawn2<-T4.Dawn2$DO.percent
T4.O2<- mutate(T4.O2,
NEP=dusk1 - dawn1,
R=dawn2 - dusk1)
#sort
T4.O2<-T4.O2 %>%
arrange(Treatment, plant.mass..g)
################ ################ ################
# combine T1 T2 T3 T4 timepoints
################ ################ ################
O2.tanks<-rbind(T1.O2,T2.O2, T3.O2, T4.O2)
cols<-c("Time.point", "Treatment", "Tank") # columns to make factors
O2.tanks[cols] <- lapply(O2.tanks[cols], factor) # make all these factors
O2.tanks$plant.mass..g<-as.numeric(O2.tanks$plant.mass..g)DO and O2%
TIME POINT 1: Change in O2% from dissolved oxygen. First, we will also run model fitting on the raw DO% data to apply the same approach for visualizing changes in oxygen across dawn-dusk-dawn measurements. We will then combine all these plots into multi panel figurs for NEP and R, and DO% for each point of measurement.
#########################################################
##################################################################
# total oxygen % plot for the 3 time points (dawn-dusk-dawn)
m1.dawn1.T1<-gam(dawn1 ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T1", data = O2.tanks, method = "REML")
m2.dawn1.T1<-gam(dawn1 ~ Treatment + s(plant.mass..g), subset = Time.point=="T1", data = O2.tanks, method = "REML")
m3.dawn1.T1<-gam(dawn1 ~ s(plant.mass..g), subset = Time.point=="T1", data = O2.tanks, method = "REML")
T1.dawn1.AIC<-AIC(m1.dawn1.T1, m2.dawn1.T1, m3.dawn1.T1)
# global with treatment best
summary(m2.dawn1.T1)
anova.gam(m2.dawn1.T1)
gam.check(m2.dawn1.T1, rep=1000)draw(m2.dawn1.T1)concrvity(m2.dawn1.T1)
par(mfrow = c(2, 2))
plot(m2.dawn1.T1, all.terms = TRUE, page=1)# model predictions
dawn1.diff.T1<-plot_difference(
m1.dawn1.T1,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
dawn1.T1.mod.plot<-
plot_smooths(
model = m2.dawn1.T1,
series = plant.mass..g,
comparison=Treatment
) +
geom_point(data=T1.O2, aes(x=plant.mass..g, y=dawn1, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
geom_line(aes(fill=Treatment, linetype=Treatment)) +
geom_hline(yintercept=0, linetype="longdash", color = "gray") +
ggtitle("T1.dawn1")+
coord_cartesian(ylim=c(-30, 150)) +
ylab(expression(paste("O"[2],"%"))) +
xlab("plant material (g)") +
theme(legend.position = "right") +
Fig.formatting
# treatment (p=0.0279) and smoothers significant (p<0.001)
####### #### Dusk 1
m1.dusk1.T1<-gam(dusk1 ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T1", data = O2.tanks, method = "REML")
m2.dusk1.T1<-gam(dusk1 ~ Treatment + s(plant.mass..g), subset = Time.point=="T1", data = O2.tanks, method = "REML")
m3.dusk1.T1<-gam(dusk1 ~ s(plant.mass..g), subset = Time.point=="T1", data = O2.tanks, method = "REML")
T1.dusk1.AIC<-AIC(m1.dusk1.T1, m2.dusk1.T1, m3.dusk1.T1)
# model with treatment and global smooth best
summary(m2.dusk1.T1)
anova.gam(m2.dusk1.T1)
gam.check(m2.dusk1.T1, rep=1000)draw(m2.dusk1.T1)concrvity(m2.dusk1.T1)
par(mfrow = c(2, 2))
plot(m2.dusk1.T1, all.terms = TRUE, page=1)# model predictions
dusk1.diff.T1<-plot_difference(
m1.dusk1.T1,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
dusk1.T1.mod.plot<-
plot_smooths(
model = m2.dusk1.T1,
series = plant.mass..g,
comparison= Treatment
) +
geom_point(data=T1.O2, aes(x=plant.mass..g, y=dusk1, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
geom_line(aes(fill=Treatment, linetype=Treatment)) +
geom_hline(yintercept=0, linetype="longdash", color = "gray") +
ggtitle("T1.dusk1")+
coord_cartesian(ylim=c(-30, 150)) +
ylab(expression(paste("O"[2],"%"))) +
xlab("plant material (g)") +
theme(legend.position = "right") +
Fig.formatting
# smoother significant for both treatments
####### #### Dawn 2
m1.dawn2.T1<-gam(dawn2 ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T1", data = O2.tanks, method = "REML")
m2.dawn2.T1<-gam(dawn2 ~ Treatment + s(plant.mass..g), subset = Time.point=="T1", data = O2.tanks, method = "REML")
m3.dawn2.T1<-gam(dawn2 ~ s(plant.mass..g), subset = Time.point=="T1", data = O2.tanks, method = "REML")
T1.dawn2.AIC<-AIC(m1.dawn2.T1, m2.dawn2.T1, m3.dawn2.T1)
# treatment and global smooth best
summary(m2.dawn2.T1)
anova.gam(m2.dawn2.T1)
gam.check(m2.dawn2.T1, rep=1000)draw(m2.dawn2.T1)concrvity(m2.dawn2.T1)
par(mfrow = c(2, 2))
plot(m2.dawn2.T1, all.terms = TRUE, page=1)# model predictions
dawn2.diff.T1<-plot_difference(
m1.dawn2.T1,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned")),
)
###########
#plot for the model output on rawdata
dawn2.T1.mod.plot<-
plot_smooths(
model = m2.dawn2.T1,
series = plant.mass..g,
comparison=Treatment
) +
geom_point(data=T1.O2, aes(x=plant.mass..g, y=dawn2, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
geom_line(aes(fill=Treatment, linetype=Treatment)) +
geom_hline(yintercept=0, linetype="longdash", color = "gray") +
ggtitle("T1.dawn2")+
coord_cartesian(ylim=c(-30, 150)) +
ylab(expression(paste("O"[2],"%"))) +
xlab("plant material (g)") +
theme(legend.position = "right") +
Fig.formatting
# smoother significant for both (p<0.001)
#### group plots
O2.T1<-plot_grid(
dawn1.T1.mod.plot+ theme(legend.position = "none"),
dusk1.T1.mod.plot+ theme(legend.position = "none"),
dawn2.T1.mod.plot+ theme(legend.position = "none"),
extract.legend,
rel_widths = c(8,8,8,3), ncol=4)TIME POINT 2: Change in O2% from dissolved oxygen
############################################################
##############################################################################
# total oxygen % plot for the 3 time points (dawn-dusk-dawn)
m1.dawn1.T2<-gam(dawn1 ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T2", data = O2.tanks, method = "REML")
m2.dawn1.T2<-gam(dawn1 ~ Treatment + s(plant.mass..g), subset = Time.point=="T2", data = O2.tanks, method = "REML")
m3.dawn1.T2<-gam(dawn1 ~ s(plant.mass..g), subset = Time.point=="T2", data = O2.tanks, method = "REML")
T2.dawn1.AIC<-AIC(m1.dawn1.T2, m2.dawn1.T2, m3.dawn1.T2)
# factor by smooth best
summary(m1.dawn1.T2)
anova.gam(m1.dawn1.T2)
gam.check(m1.dawn1.T2, rep=1000)
draw(m1.dawn1.T2)
concrvity(m1.dawn1.T2)
par(mfrow = c(2, 2))
plot(m1.dawn1.T2, all.terms = TRUE, page=1)
# model predictions
dawn1.diff.T2<-plot_difference(
m1.dawn1.T2,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
dawn1.T2.mod.plot<-
plot_smooths(
model = m1.dawn1.T2,
series = plant.mass..g,
comparison = Treatment
) +
geom_point(data=T2.O2, aes(x=plant.mass..g, y=dawn1, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
geom_hline(yintercept=0, linetype="longdash", color = "gray") +
ggtitle("T2.dawn1")+
coord_cartesian(ylim=c(-30, 150)) +
ylab(expression(paste("O"[2],"%"))) +
xlab("plant material (g)") +
theme(legend.position = "right") +
Fig.formatting
# smoothers significant (p<0.001)
####### #### Dusk 1
m1.dusk1.T2<-gam(dusk1 ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T2", data = O2.tanks, method = "REML")
m2.dusk1.T2<-gam(dusk1 ~ Treatment + s(plant.mass..g), subset = Time.point=="T2", data = O2.tanks, method = "REML")
m3.dusk1.T2<-gam(dusk1 ~ s(plant.mass..g), subset = Time.point=="T2", data = O2.tanks, method = "REML")
T2.dusk1.AIC<-AIC(m1.dusk1.T2, m2.dusk1.T2, m3.dusk1.T2)
# model with smooth by factor best
summary(m1.dusk1.T2)
anova.gam(m1.dusk1.T2)
gam.check(m1.dusk1.T2, rep=1000)
draw(m1.dusk1.T2)
concrvity(m1.dusk1.T2)
par(mfrow = c(2, 2))
plot(m1.dusk1.T2, all.terms = TRUE, page=1)
# model predictions
dusk1.diff.T2<-plot_difference(
m1.dusk1.T2,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
dusk1.T2.mod.plot<-
plot_smooths(
model = m1.dusk1.T2,
series = plant.mass..g,
comparison = Treatment
) +
geom_point(data=T2.O2, aes(x=plant.mass..g, y=dusk1, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
geom_hline(yintercept=0, linetype="longdash", color = "gray") +
ggtitle("T2.dusk1")+
coord_cartesian(ylim=c(-30, 150)) +
ylab(expression(paste("O"[2],"%"))) +
xlab("plant material (g)") +
theme(legend.position = "right") +
Fig.formatting
# smoother significant for both treatments
####### #### Dawn 2
m1.dawn2.T2<-gam(dawn2 ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T2", data = O2.tanks, method = "REML")
m2.dawn2.T2<-gam(dawn2 ~ Treatment + s(plant.mass..g), subset = Time.point=="T2", data = O2.tanks, method = "REML")
m3.dawn2.T2<-gam(dawn2 ~ s(plant.mass..g), subset = Time.point=="T2", data = O2.tanks, method = "REML")
T2.dawn2.AIC<-AIC(m1.dawn2.T2, m2.dawn2.T2, m3.dawn2.T2)
# smooth by factor best
summary(m1.dawn2.T2)
anova.gam(m1.dawn2.T2)
gam.check(m1.dawn2.T2, rep=1000)
draw(m1.dawn2.T2)
concrvity(m1.dawn2.T2)
par(mfrow = c(2, 2))
plot(m1.dawn2.T2, all.terms = TRUE, page=1)
# model predictions
dawn2.diff.T2<-plot_difference(
m1.dawn2.T2,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned")),
)
###########
#plot for the model output on rawdata
dawn2.T2.mod.plot<-
plot_smooths(
model = m1.dawn2.T2,
series = plant.mass..g,
comparison=Treatment
) +
geom_point(data=T2.O2, aes(x=plant.mass..g, y=dawn2, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
geom_hline(yintercept=0, linetype="longdash", color = "gray") +
ggtitle("T2.dawn2")+
coord_cartesian(ylim=c(-30, 150)) +
ylab(expression(paste("O"[2],"%"))) +
xlab("plant material (g)") +
theme(legend.position = "right") +
Fig.formatting
# smoother significant for both (p<0.001)
#### group plots
O2.T2<-plot_grid(
dawn1.T2.mod.plot+ theme(legend.position = "none"),
dusk1.T2.mod.plot+ theme(legend.position = "none"),
dawn2.T2.mod.plot+ theme(legend.position = "none"),
extract.legend,
rel_widths = c(8,8,8,3), ncol=4)TIME POINT 3: Change in O2% from dissolved oxygen
############################################################
##############################################################################
#### Dawn1
m1.dawn1.T3<-gam(dawn1 ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T3", data = O2.tanks, method = "REML")
m2.dawn1.T3<-gam(dawn1 ~ Treatment + s(plant.mass..g), subset = Time.point=="T3", data = O2.tanks, method = "REML")
m3.dawn1.T3<-gam(dawn1 ~ s(plant.mass..g), subset = Time.point=="T3", data = O2.tanks, method = "REML")
T3.dawn1.AIC<-AIC(m1.dawn1.T3, m2.dawn1.T3, m3.dawn1.T3)
# factor by smooth best
summary(m1.dawn1.T3)
anova.gam(m1.dawn1.T3)
gam.check(m1.dawn1.T3, rep=1000)
draw(m1.dawn1.T3)
concrvity(m1.dawn1.T3)
par(mfrow = c(2, 2))
plot(m1.dawn1.T3, all.terms = TRUE, page=1)
# model predictions
dawn1.diff.T3<-plot_difference(
m1.dawn1.T3,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
dawn1.T3.mod.plot<-
plot_smooths(
model = m1.dawn1.T3,
series = plant.mass..g,
comparison = Treatment
) +
geom_point(data=T3.O2, aes(x=plant.mass..g, y=dawn1, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
geom_hline(yintercept=0, linetype="longdash", color = "gray") +
ggtitle("T3.dawn1")+
coord_cartesian(ylim=c(-30, 150)) +
ylab(expression(paste("O"[2],"%"))) +
xlab("plant material (g)") +
theme(legend.position = "right") +
Fig.formatting
# smoothers significant (p<0.001)
####### #### Dusk 1
m1.dusk1.T3<-gam(dusk1 ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T3", data = O2.tanks, method = "REML")
m2.dusk1.T3<-gam(dusk1 ~ Treatment + s(plant.mass..g), subset = Time.point=="T3", data = O2.tanks, method = "REML")
m3.dusk1.T3<-gam(dusk1 ~ s(plant.mass..g), subset = Time.point=="T3", data = O2.tanks, method = "REML")
T3.dusk1.AIC<-AIC(m1.dusk1.T3, m2.dusk1.T3, m3.dusk1.T3)
# model with smooth by factor best
summary(m1.dusk1.T3)
anova.gam(m1.dusk1.T3)
gam.check(m1.dusk1.T3, rep=1000)
draw(m1.dusk1.T3)
concrvity(m1.dusk1.T3)
par(mfrow = c(2, 2))
plot(m1.dusk1.T3, all.terms = TRUE, page=1)
# model predictions
dusk1.diff.T3<-plot_difference(
m1.dusk1.T3,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
dusk1.T3.mod.plot<-
plot_smooths(
model = m1.dusk1.T3,
series = plant.mass..g,
comparison = Treatment
) +
geom_point(data=T3.O2, aes(x=plant.mass..g, y=dusk1, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
geom_hline(yintercept=0, linetype="longdash", color = "gray") +
ggtitle("T3.dusk1")+
coord_cartesian(ylim=c(-30, 150)) +
ylab(expression(paste("O"[2],"%"))) +
xlab("plant material (g)") +
theme(legend.position = "right") +
Fig.formatting
# no treatment effect
# smoother significant for burned
####### #### Dawn 2
m1.dawn2.T3<-gam(dawn2 ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T3", data = O2.tanks, method = "REML")
m2.dawn2.T3<-gam(dawn2 ~ Treatment + s(plant.mass..g), subset = Time.point=="T3", data = O2.tanks, method = "REML")
m3.dawn2.T3<-gam(dawn2 ~ s(plant.mass..g), subset = Time.point=="T3", data = O2.tanks, method = "REML")
T3.dawn2.AIC<-AIC(m1.dawn2.T3, m2.dawn2.T3, m3.dawn2.T3)
# smooth by factor best
summary(m1.dawn2.T3)
anova.gam(m1.dawn2.T3)
gam.check(m1.dawn2.T3, rep=1000)
draw(m1.dawn2.T3)
concrvity(m1.dawn2.T3)
par(mfrow = c(2, 2))
plot(m1.dawn2.T3, all.terms = TRUE, page=1)
# model predictions
dawn2.diff.T3<-plot_difference(
m1.dawn2.T3,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned")),
)
###########
#plot for the model output on rawdata
dawn2.T3.mod.plot<-
plot_smooths(
model = m1.dawn2.T3,
series = plant.mass..g,
comparison=Treatment
) +
geom_point(data=T3.O2, aes(x=plant.mass..g, y=dawn2, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
geom_hline(yintercept=0, linetype="longdash", color = "gray") +
ggtitle("T3.dawn2")+
coord_cartesian(ylim=c(-30, 150)) +
ylab(expression(paste("O"[2],"%"))) +
xlab("plant material (g)") +
theme(legend.position = "right") +
Fig.formatting
# global smoother significant(p<0.001)
#### group plots
O2.T3<-plot_grid(
dawn1.T3.mod.plot+ theme(legend.position = "none"),
dusk1.T3.mod.plot+ theme(legend.position = "none"),
dawn2.T3.mod.plot+ theme(legend.position = "none"),
extract.legend,
rel_widths = c(8,8,8,3), ncol=4)TIME POINT 4: Change in O2% from dissolved oxygen
############################################################
##############################################################################
# total oxygen % plot for the 3 time points (dawn-dusk-dawn)
#### Dawn1
m1.dawn1.T4<-gam(dawn1 ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T4", data = O2.tanks, method = "REML")
m2.dawn1.T4<-gam(dawn1 ~ Treatment + s(plant.mass..g), subset = Time.point=="T4", data = O2.tanks, method = "REML")
m3.dawn1.T4<-gam(dawn1 ~ s(plant.mass..g), subset = Time.point=="T4", data = O2.tanks, method = "REML")
T4.dawn1.AIC<-AIC(m1.dawn1.T4, m2.dawn1.T4, m3.dawn1.T4)
# model with global best
summary(m3.dawn1.T4)
anova.gam(m3.dawn1.T4)
gam.check(m3.dawn1.T4, rep=1000)
draw(m3.dawn1.T4)
concrvity(m3.dawn1.T4)
par(mfrow = c(2, 2))
plot(m3.dawn1.T4, all.terms = TRUE, page=1)
# model predictions
dawn1.diff.T4<-plot_difference(
m1.dawn1.T4,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
dawn1.T4.mod.plot<-
plot_smooths(
model = m3.dawn1.T4,
series = plant.mass..g,
) +
geom_point(data=T4.O2, aes(x=plant.mass..g, y=dawn1, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
geom_hline(yintercept=0, linetype="longdash", color = "gray") +
ggtitle("T4.dawn1")+
coord_cartesian(ylim=c(-30, 150)) +
ylab(expression(paste("O"[2],"%"))) +
xlab("plant material (g)") +
theme(legend.position = "right") +
Fig.formatting
# smoothers significant (p<0.001)
####### #### Dusk 1
m1.dusk1.T4<-gam(dusk1 ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T4", data = O2.tanks, method = "REML")
m2.dusk1.T4<-gam(dusk1 ~ Treatment + s(plant.mass..g), subset = Time.point=="T4", data = O2.tanks, method = "REML")
m3.dusk1.T4<-gam(dusk1 ~ s(plant.mass..g), subset = Time.point=="T4", data = O2.tanks, method = "REML")
T4.dusk1.AIC<-AIC(m1.dusk1.T4, m2.dusk1.T4, m3.dusk1.T4)
# model with smooth by factor best
summary(m1.dusk1.T4)
anova.gam(m1.dusk1.T4)
gam.check(m1.dusk1.T4, rep=1000)
draw(m1.dusk1.T4)
concrvity(m1.dusk1.T4)
par(mfrow = c(2, 2))
plot(m1.dusk1.T4, all.terms = TRUE, page=1)
# model predictions
dusk1.diff.T4<-plot_difference(
m1.dusk1.T4,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
dusk1.T4.mod.plot<-
plot_smooths(
model = m1.dusk1.T4,
series = plant.mass..g,
comparison = Treatment
) +
geom_point(data=T4.O2, aes(x=plant.mass..g, y=dusk1, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
geom_hline(yintercept=0, linetype="longdash", color = "gray") +
ggtitle("T4.dusk1")+
coord_cartesian(ylim=c(-30, 150)) +
ylab(expression(paste("O"[2],"%"))) +
xlab("plant material (g)") +
theme(legend.position = "right") +
Fig.formatting
# no treatment effect
# smoother significant for burned
####### #### Dawn 2
m1.dawn2.T4<-gam(dawn2 ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T4", data = O2.tanks, method = "REML")
m2.dawn2.T4<-gam(dawn2 ~ Treatment + s(plant.mass..g), subset = Time.point=="T4", data = O2.tanks, method = "REML")
m3.dawn2.T4<-gam(dawn2 ~ s(plant.mass..g), subset = Time.point=="T4", data = O2.tanks, method = "REML")
T4.dawn2.AIC<-AIC(m1.dawn2.T4, m2.dawn2.T4, m3.dawn2.T4)
# global smooth best
summary(m3.dawn2.T4)
anova.gam(m3.dawn2.T4)
gam.check(m3.dawn2.T4, rep=1000)
draw(m3.dawn2.T4)
concrvity(m3.dawn2.T4)
par(mfrow = c(2, 2))
plot(m3.dawn2.T4, all.terms = TRUE, page=1)
# model predictions
dawn2.diff.T4<-plot_difference(
m1.dawn2.T4,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned")),
)
###########
#plot for the model output on rawdata
dawn2.T4.mod.plot<-
plot_smooths(
model = m3.dawn2.T4,
series = plant.mass..g,
) +
geom_point(data=T4.O2, aes(x=plant.mass..g, y=dawn2, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
geom_hline(yintercept=0, linetype="longdash", color = "gray") +
ggtitle("T4.dawn2")+
coord_cartesian(ylim=c(-30, 150)) +
ylab(expression(paste("O"[2],"%"))) +
xlab("plant material (g)") +
theme(legend.position = "right") +
Fig.formatting
# global smoother significant(p<0.001)
#### group plots
O2.T4<-plot_grid(
dawn1.T4.mod.plot+ theme(legend.position = "none"),
dusk1.T4.mod.plot+ theme(legend.position = "none"),
dawn2.T4.mod.plot+ theme(legend.position = "none"),
extract.legend,
rel_widths = c(8,8,8,3), ncol=4)Combine and export all the O2 data with plot-difference and model AIC tables
#### model differences
O2.mod.diffs<-plot_grid(
dawn1.diff.T1+ theme(legend.position = "none")+ ggtitle("T1-Dawn1"),
dusk1.diff.T1+ theme(legend.position = "none")+ ggtitle("Dusk1"),
dawn2.diff.T1+ theme(legend.position = "none")+ ggtitle("Dawn2"),
dawn1.diff.T2+ theme(legend.position = "none")+ ggtitle("T2-Dawn1"),
dusk1.diff.T2+ theme(legend.position = "none")+ ggtitle("Dusk1"),
dawn2.diff.T2+ theme(legend.position = "none")+ ggtitle("Dawn2"),
dawn1.diff.T3+ theme(legend.position = "none")+ ggtitle("T3-Dawn1"),
dusk1.diff.T3+ theme(legend.position = "none")+ ggtitle("Dusk1"),
dawn2.diff.T3+ theme(legend.position = "none")+ ggtitle("Dawn2"),
dawn1.diff.T4+ theme(legend.position = "none")+ ggtitle("T4-Dawn1"),
dusk1.diff.T4+ theme(legend.position = "none")+ ggtitle("Dusk1"),
dawn2.diff.T4+ theme(legend.position = "none")+ ggtitle("Dawn2"),
rel_widths = c(8,8,8), ncol=3, nrow=4)
ggsave("figures/O2.mod.diffs.pdf", height=10, width=7)
#### model and raw data
O2.mods<-plot_grid(
O2.T1+ theme(legend.position = "none")+ ggtitle("Day-10"),
O2.T2+ theme(legend.position = "none")+ ggtitle("Day-31"),
O2.T3+ theme(legend.position = "none")+ ggtitle("Day-59"),
O2.T4+ theme(legend.position = "none")+ ggtitle("Day-89"),
rel_widths = c(8,8,8,8), ncol=1, nrow=4)
ggsave("figures/O2.mod.pdf", height=11, width=9)
#bind the AIC tables
AIC.O2<-bind_rows(T1.dawn1.AIC, T1.dusk1.AIC, T1.dawn2.AIC,
T2.dawn1.AIC, T2.dusk1.AIC, T2.dawn2.AIC,
T3.dawn1.AIC, T3.dusk1.AIC, T3.dawn2.AIC,
T4.dawn1.AIC, T4.dusk1.AIC, T4.dawn2.AIC)
# make a model column
mod.rep12<-rep(c("~Treatment + s(plant.mass..g, by= Treatment)",
"~Treatment + s(plant.mass..g)",
"~s(plant.mass..g)"), times=12)
mod.O2.df<- data.frame(mod.rep12)
#bind table
AIC.O2.mod<-cbind(mod.O2.df, AIC.O2)
write.csv(AIC.O2.mod, "output/AIC models/AIC.O2.mod.csv")NEP and R
Generate dataframes for NEP and R change in O2. We will use NEP and R, run gam model fits, and produce individual figures for each time point.
First, we will use NEP for productivity measurements.
####### Time 1
m1.NEP.T1<-gam(NEP ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T1", data = O2.tanks, method = "REML")
m2.NEP.T1<-gam(NEP ~ Treatment + s(plant.mass..g), subset = Time.point=="T1", data = O2.tanks, method = "REML")
m3.NEP.T1<-gam(NEP ~ s(plant.mass..g), subset = Time.point=="T1", data = O2.tanks, method = "REML")
T1.NEP.AIC<-AIC(m1.NEP.T1, m2.NEP.T1, m3.NEP.T1)
# model with plot smooth by factor not different from reduced model, go with smooth by factor
summary(m2.NEP.T1)
anova.gam(m2.NEP.T1)
gam.check(m2.NEP.T1, rep=1000)
draw(m2.NEP.T1)
concrvity(m2.NEP.T1)
par(mfrow = c(2, 2))
plot(m2.NEP.T1, all.terms = TRUE, page=1)
#### see this https://cran.r-project.org/web/packages/tidymv/vignettes/plot-smooths.html
# The difference smooth is difference between the smooths of two conditions (two levels in a factor).
# Portions of the difference smooth confidence interval that do not include 0 are shaded in red.
# model predictions
NEP.diff.T1<-plot_difference(
m2.NEP.T1,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
NEP.T1.mod.plot<-
plot_smooths(
model = m2.NEP.T1,
series = plant.mass..g,
comparison = Treatment
) +
geom_point(data=T1.O2, aes(x=plant.mass..g, y=NEP, color=Treatment)) +
geom_line(aes(fill=Treatment, linetype=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
coord_cartesian(ylim=c(-20, 50)) +
geom_hline(yintercept=0, linetype="longdash", color = "gray") +
ylab(expression(paste("Net Production (", Delta, "O"[2],"%)"))) +
theme(legend.position = "right") +
Fig.formatting
# no treatment effect (p=0.110), smoothers significant (p<0.006)
####### Time 2
m1.NEP.T2<-gam(NEP ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T2", data = O2.tanks, method = "REML")
m2.NEP.T2<-gam(NEP ~ Treatment + s(plant.mass..g), subset = Time.point=="T2", data = O2.tanks, method = "REML")
m3.NEP.T2<-gam(NEP ~ s(plant.mass..g), subset = Time.point=="T2", data = O2.tanks, method = "REML")
T2.NEP.AIC<-AIC(m1.NEP.T2, m2.NEP.T2, m3.NEP.T2)
# model with plot smooth by factor not different from reduced model, go with smooth by factor
summary(m2.NEP.T2)
anova.gam(m2.NEP.T2)
gam.check(m2.NEP.T2, rep=1000)
draw(m2.NEP.T2)
concrvity(m2.NEP.T2)
par(mfrow = c(2, 2))
plot(m2.NEP.T2, all.terms = TRUE, page=1)
## plot for the model output on rawdata
# model predictions
NEP.diff.T2<-plot_difference(
m2.NEP.T2,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
NEP.T2.mod.plot<-
plot_smooths(
model = m2.NEP.T2,
series = plant.mass..g,
comparison = Treatment
) +
geom_point(data=T2.O2, aes(x=plant.mass..g, y=NEP, color=Treatment)) +
geom_line(aes(fill=Treatment, linetype=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
coord_cartesian(ylim=c(-20, 50)) +
geom_hline(yintercept=0, linetype="longdash", color = "gray") +
ylab(expression(paste("Net Production (", Delta, "O"[2],"%)"))) +
theme(legend.position = "right") +
Fig.formatting
# treatment effect (p=0.007)
# smoother significant for burned (p=0.002) but not unburned (p=0.326)
####### Time 3
m1.NEP.T3<-gam(NEP ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T3", data = O2.tanks, method = "REML")
m2.NEP.T3<-gam(NEP ~ Treatment + s(plant.mass..g), subset = Time.point=="T3", data = O2.tanks, method = "REML")
m3.NEP.T3<-gam(NEP ~ s(plant.mass..g), subset = Time.point=="T3", data = O2.tanks, method = "REML")
T3.NEP.AIC<-AIC(m1.NEP.T3, m2.NEP.T3, m3.NEP.T3)
# model with plot smooth by factor not different from reduced model, go with smooth by factor
summary(m2.NEP.T3)
anova.gam(m2.NEP.T3)
gam.check(m2.NEP.T3, rep=1000)
draw(m2.NEP.T3)
concrvity(m2.NEP.T3)
par(mfrow = c(2, 2))
plot(m2.NEP.T3, all.terms = TRUE, page=1)
# model predictions
NEP.diff.T3<-plot_difference(
m2.NEP.T3,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
#plot for the model output on rawdata
NEP.T3.mod.plot<-
plot_smooths(
model = m2.NEP.T3,
series = plant.mass..g,
comparison = Treatment
) +
geom_point(data=T3.O2, aes(x=plant.mass..g, y=NEP, color=Treatment)) +
geom_line(aes(fill=Treatment, linetype=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
coord_cartesian(ylim=c(-20, 50)) +
geom_hline(yintercept=0, linetype="longdash", color = "gray") +
ylab(expression(paste("Net Production (", Delta, "O"[2],"%)"))) +
theme(legend.position = "right") +
Fig.formatting
# treatment effect (p=0.009)
# smoother significant for burned (p<0.001) but not unburned (p=0.053)
####### Time 4
m1.NEP.T4<-gam(NEP ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T4", data = O2.tanks, method = "REML")
m2.NEP.T4<-gam(NEP ~ Treatment + s(plant.mass..g), subset = Time.point=="T4", data = O2.tanks, method = "REML")
m3.NEP.T4<-gam(NEP ~ s(plant.mass..g), subset = Time.point=="T4", data = O2.tanks, method = "REML")
T4.NEP.AIC<-AIC(m1.NEP.T4, m2.NEP.T4, m3.NEP.T4)
# model with plot smooth by factor not different from reduced model, go with smooth by factor
summary(m1.NEP.T4)
anova.gam(m1.NEP.T4)
gam.check(m1.NEP.T4, rep=1000)
draw(m1.NEP.T4)
concrvity(m1.NEP.T4)
par(mfrow = c(2, 2))
plot(m1.NEP.T4, all.terms = TRUE, page=1)
# model predictions
NEP.diff.T4<-plot_difference(
m1.NEP.T4,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
#plot for the model output on rawdata
NEP.T4.mod.plot<-
plot_smooths(
model = m1.NEP.T4,
series = plant.mass..g,
comparison = Treatment
) +
geom_point(data=T4.O2, aes(x=plant.mass..g, y=NEP, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
coord_cartesian(ylim=c(-20, 50)) +
geom_hline(yintercept=0, linetype="longdash", color = "gray") +
ylab(expression(paste("Net Production (", Delta, "O"[2],"%)"))) +
theme(legend.position = "right") +
Fig.formatting
# no treatment effect (p=0.118)
# smoother significant for burned (p=0.020) but not unburned (p=0.327)
mod.RNEP<-rep(c("~Treatment + s(plant.mass..g, by= Treatment)",
"~Treatment + s(plant.mass..g)",
"~s(plant.mass..g)"), times=4)
mod.RNEP.df<- data.frame(mod.RNEP)
AIC.NEP<-bind_rows(T1.NEP.AIC, T2.NEP.AIC, T3.NEP.AIC, T4.NEP.AIC)
AIC.NEP.mod<-cbind(mod.RNEP.df, AIC.NEP)
write.csv(AIC.NEP.mod, "output/AIC models/AIC.NEP.mod.csv")Table: Results for Time-1 NEP
anova.gam(m1.NEP.T1)##
## Family: gaussian
## Link function: identity
##
## Formula:
## NEP ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric Terms:
## df F p-value
## Treatment 1 2.756 0.11
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 2.478 3.046 13.44 2.3e-05
## s(plant.mass..g):Treatmentunburned 1.690 2.090 6.36 0.00585Table: Results for Time-2 NEP
anova.gam(m1.NEP.T2)##
## Family: gaussian
## Link function: identity
##
## Formula:
## NEP ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric Terms:
## df F p-value
## Treatment 1 8.479 0.00745
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 1.967 2.431 7.392 0.00227
## s(plant.mass..g):Treatmentunburned 1.000 1.000 1.002 0.32638Table: Results for Time-3 NEP
anova.gam(m1.NEP.T3)##
## Family: gaussian
## Link function: identity
##
## Formula:
## NEP ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric Terms:
## df F p-value
## Treatment 1 8.118 0.00948
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 3.414 4.166 8.236 0.000334
## s(plant.mass..g):Treatmentunburned 3.124 3.821 2.940 0.053008Table: Results for Time-4 NEP
anova.gam(m1.NEP.T4)##
## Family: gaussian
## Link function: identity
##
## Formula:
## NEP ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric Terms:
## df F p-value
## Treatment 1 2.62 0.118
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 2.757 3.382 3.717 0.0203
## s(plant.mass..g):Treatmentunburned 1.000 1.000 1.002 0.3268Now, we will go through Respiration models and individual plots.
####### Time 1
m1.R.T1<-gam(R ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T1", data = O2.tanks, method = "REML")
m2.R.T1<-gam(R ~ Treatment + s(plant.mass..g), subset = Time.point=="T1", data = O2.tanks, method = "REML")
m3.R.T1<-gam(R ~ s(plant.mass..g), subset = Time.point=="T1", data = O2.tanks, method = "REML")
T1.R.AIC<-AIC(m1.R.T1, m2.R.T1, m3.R.T1)
# model with global best
summary(m3.R.T1)
anova.gam(m3.R.T1)
gam.check(m3.R.T1, rep=1000)draw(m3.R.T1)concrvity(m3.R.T1)
par(mfrow = c(2, 2))
plot(m3.R.T1, all.terms = TRUE, page=1)# model predictions
R.diff.T1<-plot_difference(
m1.R.T1,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
R.T1.mod.plot<-
plot_smooths(
model = m3.R.T1,
series = plant.mass..g,
) +
geom_point(data=T1.O2, aes(x=plant.mass..g, y=R, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
coord_cartesian(ylim=c(-40, 10)) +
geom_hline(yintercept=0, linetype="longdash", color = "gray") +
ylab(expression(paste("Net Respiration (", Delta, "O"[2],"%)"))) +
theme(legend.position = "right") +
Fig.formatting
# no treatment effect (p=0.229), smoothers significant (p<0.001)
####### Time 2
m1.R.T2<-gam(R ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T2", data = O2.tanks, method = "REML")
m2.R.T2<-gam(R ~ Treatment + s(plant.mass..g), subset = Time.point=="T2", data = O2.tanks, method = "REML")
m3.R.T2<-gam(R ~ s(plant.mass..g), subset = Time.point=="T2", data = O2.tanks, method = "REML")
T2.R.AIC<-AIC(m1.R.T2, m2.R.T2, m3.R.T2)
# model with global + treatment best
summary(m2.R.T2)
anova.gam(m2.R.T2)
gam.check(m2.R.T2, rep=1000)draw(m2.R.T2)concrvity(m2.R.T2)
par(mfrow = c(2, 2))
plot(m2.R.T2, all.terms = TRUE, page=1)# model predictions
R.diff.T2<-plot_difference(
m2.R.T2,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
R.T2.mod.plot<-
plot_smooths(
model = m2.R.T2,
series = plant.mass..g,
comparison= Treatment,
) +
geom_point(data=T2.O2, aes(x=plant.mass..g, y=R, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
geom_line(aes(fill=Treatment, linetype=Treatment)) +
coord_cartesian(ylim=c(-40, 10)) +
geom_hline(yintercept=0, linetype="longdash", color = "gray") +
ylab(expression(paste("Net Respiration (", Delta, "O"[2],"%)"))) +
theme(legend.position = "right") +
Fig.formatting
# slight treatment effect (p=0.085)
# smoother significant for both burned and unburned (p<0.008)
####### Time 3
m1.R.T3<-gam(R ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T3", data = O2.tanks, method = "REML")
m2.R.T3<-gam(R ~ Treatment + s(plant.mass..g), subset = Time.point=="T3", data = O2.tanks, method = "REML")
m3.R.T3<-gam(R ~ s(plant.mass..g), subset = Time.point=="T3", data = O2.tanks, method = "REML")
T3.R.AIC<-AIC(m1.R.T3, m2.R.T3, m3.R.T3)
# model smmoth by factor best
summary(m1.R.T3)
anova.gam(m1.R.T3)
gam.check(m1.R.T3, rep=1000)draw(m1.R.T3)concrvity(m1.R.T3)
par(mfrow = c(2, 2))
plot(m1.R.T3, all.terms = TRUE, page=1)# model predictions
R.diff.T3<-plot_difference(
m1.R.T3,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
R.T3.mod.plot<-
plot_smooths(
model = m1.R.T3,
series = plant.mass..g,
comparison = Treatment
) +
geom_point(data=T3.O2, aes(x=plant.mass..g, y=R, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
coord_cartesian(ylim=c(-40, 10)) +
geom_hline(yintercept=0, linetype="longdash", color = "gray") +
ylab(expression(paste("Net Respiration (", Delta, "O"[2],"%)"))) +
theme(legend.position = "right") +
Fig.formatting
# treatment effect (p=0.027)
# smoother significant for burned and unnburned (p<0.001)
####### Time 4
m1.R.T4<-gam(R ~ Treatment + s(plant.mass..g, by= Treatment), subset = Time.point=="T4", data = O2.tanks, method = "REML")
m2.R.T4<-gam(R ~ Treatment + s(plant.mass..g), subset = Time.point=="T4", data = O2.tanks, method = "REML")
m3.R.T4<-gam(R ~ s(plant.mass..g), subset = Time.point=="T4", data = O2.tanks, method = "REML")
T4.R.AIC<-AIC(m1.R.T4, m2.R.T4, m3.R.T4)
# model with global + treatment best
summary(m1.R.T4)
anova.gam(m1.R.T4)
gam.check(m1.R.T4, rep=1000)draw(m1.R.T4)concrvity(m1.R.T4)
par(mfrow = c(2, 2))
plot(m1.R.T4, all.terms = TRUE, page=1)# model predictions
R.diff.T4<-plot_difference(
m1.R.T4,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
R.T4.mod.plot<-
plot_smooths(
model = m1.R.T4,
series = plant.mass..g,
comparison = Treatment
) +
geom_point(data=T4.O2, aes(x=plant.mass..g, y=R, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
coord_cartesian(ylim=c(-40, 10)) +
geom_hline(yintercept=0, linetype="longdash", color = "gray") +
ylab(expression(paste("Net Respiration (", Delta, "O"[2],"%)"))) +
theme(legend.position = "right") +
Fig.formatting
# no treatment effect (p=0.078)
# smoother significant for burned (p=0.004) but not unburned (p=0.965)
R.mod.plot<-plot_grid(
R.T1.mod.plot+ theme(legend.position = "none")+ ggtitle("Day-10"),
R.T2.mod.plot+ theme(legend.position = "none")+ ggtitle("Day-31"),
R.T3.mod.plot+ theme(legend.position = "none")+ ggtitle("Day-59"),
R.T4.mod.plot+ theme(legend.position = "none")+ ggtitle("Day-89"), extract.legend,
rel_widths = c(8,8,8,8,3), ncol=5)
#ggsave("figures/R.mod.plot.long.pdf", height=6, width=12)
#### model differences
R.mod.diffs<-plot_grid(
R.diff.T1+ theme(legend.position = "none")+ ggtitle("R-Day-10"),
R.diff.T2+ theme(legend.position = "none")+ ggtitle("Day-31"),
R.diff.T3+ theme(legend.position = "none")+ ggtitle("Day-59"),
R.diff.T4+ theme(legend.position = "none")+ ggtitle("Day-89"),
rel_widths = c(8,8,8,8), ncol=4)
#ggsave("figures/R.mod.diffs.pdf", height=3, width=10)
AIC.R<-bind_rows(T1.R.AIC, T2.R.AIC, T3.R.AIC, T4.R.AIC)
AIC.R.mod<-cbind(mod.RNEP.df, AIC.R)
write.csv(AIC.R.mod, "output/AIC models/AIC.R.mod.csv")Table: Results for Time-1 R
anova.gam(m3.R.T1)##
## Family: gaussian
## Link function: identity
##
## Formula:
## R ~ s(plant.mass..g)
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g) 2.520 3.097 12.8 2.32e-05Table: Results for Time-2 R
anova.gam(m2.R.T2)##
## Family: gaussian
## Link function: identity
##
## Formula:
## R ~ Treatment + s(plant.mass..g)
##
## Parametric Terms:
## df F p-value
## Treatment 1 6.443 0.0186
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g) 5.710 6.758 10 1.52e-05Table: Results for Time-3 R
anova.gam(m1.R.T3)##
## Family: gaussian
## Link function: identity
##
## Formula:
## R ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric Terms:
## df F p-value
## Treatment 1 5.669 0.0268
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 3.762 4.576 13.144 1.02e-05
## s(plant.mass..g):Treatmentunburned 3.274 4.000 7.775 0.000523Table: Results for Time-4 R
anova.gam(m1.R.T4)##
## Family: gaussian
## Link function: identity
##
## Formula:
## R ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric Terms:
## df F p-value
## Treatment 1 3.38 0.0784
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 2.927 3.587 5.293 0.00408
## s(plant.mass..g):Treatmentunburned 1.000 1.000 0.002 0.96531Compiled NEP-R plots with model fits
NEP.R.alltimes.long<-plot_grid(
NEP.T1.mod.plot+ theme(legend.position = "none"),
NEP.T2.mod.plot+ theme(legend.position = "none"),
NEP.T3.mod.plot+ theme(legend.position = "none"),
NEP.T4.mod.plot+ theme(legend.position = "none"), extract.legend,
R.T1.mod.plot+ theme(legend.position = "none"),
R.T2.mod.plot+ theme(legend.position = "none"),
R.T3.mod.plot+ theme(legend.position = "none"),
R.T4.mod.plot+ theme(legend.position = "none"), extract.legend,
rel_widths = c(8,8,8,8,3, 8,8,8,8,3), ncol=5)
ggsave("figures/NEP.R.alltimes.long.pdf", height=7, width=12)
NEP.R.alltimes.long
Model differences between the two factor-smoothers
NEP.R.mod.diffs<-plot_grid(
NEP.diff.T1+ theme(legend.position = "none"),
NEP.diff.T2+ theme(legend.position = "none"),
NEP.diff.T3+ theme(legend.position = "none"),
NEP.diff.T4+ theme(legend.position = "none"),
R.diff.T1+ theme(legend.position = "none"),
R.diff.T2+ theme(legend.position = "none"),
R.diff.T3+ theme(legend.position = "none"),
R.diff.T4+ theme(legend.position = "none"),
rel_widths = c(8,8,8,8,8,8,8,8), ncol=4)
ggsave("figures/NEP.R.mod.diffs.alt.pdf", height=7, width=12)
NEP.R.mod.diffs
Isotopes
Isotopes and C:N for starting materials used in the experiment and plankton fractions sampled at time 1 and time 2.
######### Time 1 and Time 2
topes<-read.csv("data/Isotopes/Pyro_isotopes.csv")
topes$C.N <-(topes$Total.C..ug/12)/(topes$Total.N..ug/14) # C mol : N mol
cols<-c("Time.point", "Treatment", "Type", "Tank") # columns to make factors
topes[cols] <- lapply(topes[cols], factor) # make all these factors
##### make data frames
# treatment data df
topes.trt<-topes[(topes$Treatment=="burned" | topes$Treatment=="unburned"),]
topes.trt<-droplevels(topes.trt)
topes.trt$Type<-factor(topes.trt$Type,
levels=c("plankton", "POM"))
# control and start plant materials df
topes.controls<-topes[!(topes$Treatment=="burned" | topes$Treatment=="unburned"),]Total C vs total N for plankton at Time1 and Time2. The model here is a linear regression
############# C vs N plot
#plot for the model output on rawdata
CvN.mod1.T1<-lm(Total.N..ug~Total.C..ug*Treatment, data=topes.trt[(topes.trt$Time.point=="T1"),])
CvN.mod2.T1<-lm(Total.N..ug~Total.C..ug + Treatment, data=topes.trt[(topes.trt$Time.point=="T1"),])
anova(CvN.mod1.T1,CvN.mod2.T1) # no diff, go with simple model 'CvN.mod2.T1'## Analysis of Variance Table
##
## Model 1: Total.N..ug ~ Total.C..ug * Treatment
## Model 2: Total.N..ug ~ Total.C..ug + Treatment
## Res.Df RSS Df Sum of Sq F Pr(>F)
## 1 56 22978
## 2 57 23284 -1 -305.75 0.7451 0.3917Anova(CvN.mod1.T1, type=3)## Anova Table (Type III tests)
##
## Response: Total.N..ug
## Sum Sq Df F value Pr(>F)
## (Intercept) 1590 1 3.8744 0.05399 .
## Total.C..ug 101937 1 248.4283 < 2e-16 ***
## Treatment 15 1 0.0368 0.84859
## Total.C..ug:Treatment 306 1 0.7451 0.39170
## Residuals 22978 56
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1CvN.T1.mod.plot<-
ggplot(data=topes.trt[(topes.trt$Time.point=="T1"),],
aes(x=Total.C..ug, y=Total.N..ug,
color=Treatment, # for the points
fill=Treatment, # for the SE
linetype=Treatment)) + #for the regression lines
geom_point(aes(shape=Type))+
geom_smooth(method = "lm", alpha =0.2, size=0.5)+
scale_color_manual(values = c("brown1", "mediumseagreen"))+
ggtitle("Time-1") +
scale_shape_manual(name="Plankton", values = c(17, 16),
labels = c(expression(paste("< 63"~mu,"m")),
expression(paste("> 63"~mu,"m")))) +
ylim(c(0,600)) +
xlim(c(0,3200)) +
ylab("Total N (ug)") +
xlab("Total C (ug)") +
Fig.formatting
#############################
### Time 2 C vs. N
#plot for the model output on rawdata
CvN.mod1.T2<-lm(Total.N..ug~Total.C..ug*Treatment, data=topes.trt[(topes.trt$Time.point=="T2"),])
CvN.mod2.T2<-lm(Total.N..ug~Total.C..ug + Treatment, data=topes.trt[(topes.trt$Time.point=="T2"),])
anova(CvN.mod1.T2,CvN.mod2.T2) # no diff, go with simple model 'CvN.mod2.T2'## Analysis of Variance Table
##
## Model 1: Total.N..ug ~ Total.C..ug * Treatment
## Model 2: Total.N..ug ~ Total.C..ug + Treatment
## Res.Df RSS Df Sum of Sq F Pr(>F)
## 1 56 14419
## 2 57 14732 -1 -312.27 1.2128 0.2755Anova(CvN.mod2.T2, type=2)## Anova Table (Type II tests)
##
## Response: Total.N..ug
## Sum Sq Df F value Pr(>F)
## Total.C..ug 204416 1 790.9323 <2e-16 ***
## Treatment 194 1 0.7492 0.3904
## Residuals 14732 57
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1CvN.T2.mod.plot<-
ggplot(data=topes.trt[(topes.trt$Time.point=="T2"),],
aes(x=Total.C..ug, y=Total.N..ug,
color=Treatment, # for the points
fill=Treatment, # for the SE
linetype=Treatment)) + #for the regression lines
geom_point(aes(shape=Type))+
geom_smooth(method = "lm", alpha =0.2, size=0.5)+
scale_color_manual(values = c("brown1", "mediumseagreen"))+
ggtitle("Time-2") +
scale_shape_manual(name="Plankton", values = c(17, 16),
labels = c(expression(paste("> 63"~mu,"m")),
expression(paste("< 63"~mu,"m")))) +
ylim(c(0,250)) +
xlim(c(0,1250)) +
ylab("Total N (ug)") +
xlab("Total C (ug)") +
Fig.formatting C:N vs. plant material models and plot, fit with GAMs Test if the C:N was consistent across time, or what factors may influence C:N. This is important to verify assumptions. We are using 15N transer between trophic levels, assuming our measurements of efficiency in nitrogen transfer reflect carbon and eRgy transfer between trophic levels.
############# all plankton T1 nd T2
m1.CN <- gam(C.N ~ Treatment + Type + Time.point +
s(plant.mass..g, by=Treatment),
data=topes.trt, method="REML", family="gaussian")
m2.CN <- gam(C.N ~ Treatment + Type + Time.point +
s(plant.mass..g),
data=topes.trt, method="REML", family="gaussian")
m3.CN <- gam(C.N ~
s(plant.mass..g),
data=topes.trt, method="REML", family="gaussian")
m4.CN <- gam(C.N ~ Treatment + Type +
s(plant.mass..g, by=Time.point),
data=topes.trt, method="REML", family="gaussian")
m5.CN <- gam(C.N ~ Treatment + Type +
s(plant.mass..g, by=Treatment),
data=topes.trt, method="REML", family="gaussian")
#best model here
m6.CN <- gam(C.N ~ Type +
s(plant.mass..g, by=Treatment),
data=topes.trt, method="REML", family="gaussian")
m7.CN <- gam(C.N ~ Treatment +
s(plant.mass..g, by=Type),
data=topes.trt, method="REML", family="gaussian")
m8.CN <- gam(C.N ~ Type +
s(plant.mass..g, by=Type),
data=topes.trt, method="REML", family="gaussian")
AIC.CN<-AIC(m1.CN, m2.CN, m3.CN, m4.CN, m5.CN, m6.CN, m7.CN, m8.CN)
## additive model best fit, but no treatment or type effect
summary(m6.CN)
anova.gam(m6.CN)
gam.check(m6.CN, rep=1000)draw(m6.CN)concrvity(m6.CN)
par(mfrow = c(1, 2))
plot(m6.CN, all.terms = TRUE, page=1)# model for smoothing
msmooth.CN<-gam(C.N ~ Type +
s(plant.mass..g, by=Type),
data=topes.trt, method="REML", family="gaussian")
# model predictions
CN.diff<-plot_difference(
m1.CN,
series = plant.mass..g,
difference = list(Type = c("plankton", "POM"))
)
###########
#plot for the model output on rawdata
CN.mod.plot.timepooled<-
plot_smooths(
model = msmooth.CN,
series = plant.mass..g,
comparison = Type) +
theme(legend.position = "none") +
geom_point(data=topes.trt,
aes(x=plant.mass..g, y=C.N, color=Type, fill=Type)) +
scale_fill_manual(values = c("deepskyblue4", "darkseagreen"), guide='none') +
scale_color_manual(name="Plankton", values = c("deepskyblue4", "darkseagreen"),
labels = c(expression(paste("> 63"~mu,"m")),
expression(paste("< 63"~mu,"m")))) +
theme(legend.position = "right") +
ggtitle("Days 10 and 31") +
coord_cartesian(ylim=c(0, 20)) +
ylab("C:N") +
xlab("plant material (g)") +
Fig.formatting
CN.mod.plot.timepooledggsave("figures/CN.mod.plot.timepooled.pdf", height=4, width=5, encod="MacRoman")
#####
# linear model approach
CN.all.mod<-lm(C.N~ Treatment+Type+Time.point, data=topes.trt, na.action=na.exclude)
print(Anova(CN.all.mod, type=2), digits=5)
posthoc<-emmeans(CN.all.mod, ~Type)
multcomp::cld(posthoc, Letters=letters)
# All time C.N boxplot
CNbox.all.time<-ggplot(topes.trt, aes(x=Treatment, y=C.N, fill=Type)) +
geom_boxplot() +
geom_point(pch = 21, position = position_jitterdodge(), alpha=0.6) +
scale_fill_manual(name="Plankton", values = c("deepskyblue4", "darkseagreen"),
labels = c(expression(paste("> 63"~mu,"m")),
expression(paste("< 63"~mu,"m")))) +
coord_cartesian(ylim=c(0, 20)) +
ylab("C:N") +
Fig.formatting
CNbox.all.timeggsave("figures/CNbox.all.time.pdf", height=4, width=5, encod="MacRoman")############# all plankton T1
m1.T1.CN <- gam(C.N ~ Treatment + Type +
s(plant.mass..g, by=Treatment),
subset = Time.point=="T1", data=topes.trt, method="REML", family="gaussian")
m2.T1.CN <- gam(C.N ~ Treatment + Type +
s(plant.mass..g),
subset = Time.point=="T1", data=topes.trt, method="REML", family="gaussian")
m3.T1.CN <- gam(C.N ~
s(plant.mass..g),
subset = Time.point=="T1", data=topes.trt, method="REML", family="gaussian")
AIC.CN.T1<-AIC(m1.T1.CN, m2.T1.CN, m3.T1.CN)
## additive model best fit, but no treatment or type effect
summary(m1.T1.CN)
anova.gam(m1.T1.CN)
gam.check(m1.T1.CN, rep=1000)draw(m1.T1.CN)concrvity(m1.T1.CN)
par(mfrow = c(1, 2))
plot(m1.T1.CN, all.terms = TRUE, page=1)# model for smoothing
msmooth.T1.CN<- gam(C.N ~ Treatment +
s(plant.mass..g, by=Treatment),
subset = Time.point=="T1", data=topes.trt, method="REML", family="gaussian")
# model predictions
CN.diff.T1<-plot_difference(
m1.T1.CN,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
CN.T1.mod.plot<-
plot_smooths(
model = m1.T1.CN,
series = plant.mass..g,
comparison = Treatment
) + theme(legend.position = "none") +
geom_point(data=topes.trt[(topes.trt$Time.point=="T1"),],
aes(x=plant.mass..g, y=C.N, color=Treatment, shape=Type)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
theme(legend.position = "right") +
ggtitle("Time-1") +
coord_cartesian(ylim=c(0, 20)) +
scale_shape_manual(name="Plankton", values = c(17, 16),
labels = c(expression(paste("> 63"~mu,"m")),
expression(paste("< 63"~mu,"m")))) +
ylab("C:N") +
xlab("plant material (g)") +
Fig.formatting
# no effect of type or treatment
# smoother significant for both
############# all plankton T2
m1.T2.CN <- gam(C.N ~ Treatment + Type +
s(plant.mass..g, by=Treatment),
subset = Time.point=="T2", data=topes.trt, method="REML", family="gaussian")
m2.T2.CN <- gam(C.N ~ Treatment + Type +
s(plant.mass..g),
subset = Time.point=="T2", data=topes.trt, method="REML", family="gaussian")
m3.T2.CN <- gam(C.N ~
s(plant.mass..g),
subset = Time.point=="T2", data=topes.trt, method="REML", family="gaussian")
AIC.CN.T2<-AIC(m1.T2.CN, m2.T2.CN, m3.T2.CN)
## additive model best fit, but no treatment or type effect
summary(m1.T2.CN)
anova.gam(m1.T2.CN)
gam.check(m2.T2.CN, rep=1000)draw(m1.T2.CN)concrvity(m1.T2.CN)
par(mfrow = c(1, 2))
plot(m1.T2.CN, all.terms = TRUE, page=1) ###########
# model for smoothing
msmooth.T2.CN<- gam(C.N ~ Treatment +
s(plant.mass..g, by=Treatment),
subset = Time.point=="T2", data=topes.trt, method="REML", family="gaussian")
# model predictions
CN.diff.T2<-plot_difference(
msmooth.T2.CN,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
#plot for the model output on rawdata
CN.T2.mod.plot<-
plot_smooths(
model = msmooth.T2.CN,
series = plant.mass..g,
comparison = Treatment
) + theme(legend.position = "none") +
geom_point(data=topes.trt[(topes.trt$Time.point=="T2"),],
aes(x=plant.mass..g, y=C.N, color=Treatment, shape=Type)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
theme(legend.position = "right") +
ggtitle("Time-2") +
coord_cartesian(ylim=c(0, 20)) +
scale_shape_manual(name="Plankton", values = c(17, 16),
labels = c(expression(paste("> 63"~mu,"m")),
expression(paste("< 63"~mu,"m")))) +
ylab("C:N") +
xlab("plant material (g)") +
Fig.formatting
# no effect of treatment, but type important (higher in POM)
# smoother significant for bothCompile CvsN and C:N vs. plant material plots and models
## AIC table
mod.CN<-rep(c("Treatment + Type + s(plant.mass..g, by=Treatment)",
"Treatment + Type + s(plant.mass..g)",
"s(plant.mass..g)"), times=2)
mod.CN.df<- data.frame(mod.CN)
AIC.CN<-bind_rows(AIC.CN.T1, AIC.CN.T2)
AIC.CN.mod<-cbind(mod.CN.df, AIC.CN)
write.csv(AIC.CN.mod, "output/AIC models/AIC.CN.mod.csv")
### get legend
extract.legend.C.N <- get_legend(
# create some space to the left of the legend
CN.T2.mod.plot + theme(legend.box.margin = margin(0, 0, 0, 10)))
## combine plots
isoCNplot<- plot_grid(CvN.T1.mod.plot + theme(legend.position = "none"),
CvN.T2.mod.plot + theme(legend.position = "none"),
CN.T1.mod.plot + theme(legend.position = "none"),
CN.T2.mod.plot + theme(legend.position = "none"),
extract.legend.C.N, rel_widths = c(8,8,8,8,3), ncol=5)
ggsave("figures/isoCNplots.pdf", height=6, width=12, encod="MacRoman")
## combine model diffs
isoCNplot.diff<- plot_grid(CN.diff.T1 + theme(legend.position = "none"),
CN.diff.T2 + theme(legend.position = "none"),
rel_widths = c(8,8), ncol=2)
ggsave("figures/isoCN.moddiff.pdf", height=5, width=8, encod="MacRoman")The “controls” are the tin blanks, plankton, and starting materials
########## ########## ##########
# run some stats to see how the control material differs from each other
# make a burning treatment
topes.controls$Burn.Trt=as.factor(word(topes.controls$Type, -1, sep="[.]"))
# make a plant name
topes.controls$Plant=as.factor(word(topes.controls$Type, 1, sep="[.]"))
### test some models on controls
# remove blanks and plankton, keeping only plants
topes.controls.plants<-topes.controls[!(topes.controls$Plant=="blank" |
topes.controls$Plant=="plankton" ),]
topes.controls.plants$Plant<-droplevels(topes.controls.plants$Plant)
# keep only non-enriched samples (remove sage)
topes.controls.non.enrich<-topes.controls[!(topes.controls$Plant=="blank" |
topes.controls$Plant=="sage" ),]
topes.controls.non.enrich$Plant<-droplevels(topes.controls.non.enrich$Plant)
######## test plant species differences
mwu(topes.controls.plants, d15N, Plant)##
## # Mann-Whitney-U-Test
##
## Groups 1 = sage (n = 18) | 2 = willow (n = 8):
## U = 315.000, W = 144.000, p < .001, Z = 4.000
## effect-size r = 0.784
## rank-mean(1) = 17.50
## rank-mean(2) = 4.50# d15N sage and willow differ (p<0.001)
mwu(topes.controls.plants, C.N, Plant) ##
## # Mann-Whitney-U-Test
##
## Groups 1 = sage (n = 16) | 2 = willow (n = 8):
## U = 160.000, W = 24.000, p = 0.014, Z = -2.449
## effect-size r = 0.500
## rank-mean(1) = 10.00
## rank-mean(2) = 17.50# C.N sage and willow differ (p=0.013)
par(mfrow=c(1,2))
boxplot(d15N~Plant, data=topes.controls.plants)
boxplot(C.N~Plant, data=topes.controls.plants)
######## test difference between willow and plankton
mwu(topes.controls.non.enrich, d15N, Plant)##
## # Mann-Whitney-U-Test
##
## Groups 1 = plankton (n = 7) | 2 = willow (n = 8):
## U = 28.000, W = 0.000, p = 0.001, Z = -3.240
## effect-size r = 0.837
## rank-mean(1) = 4.00
## rank-mean(2) = 11.50 # d15N plankton and willow differ (p<0.001)
mwu(topes.controls.non.enrich, C.N, Plant) # C:N plankton and willow differ (p<0.001)##
## # Mann-Whitney-U-Test
##
## Groups 1 = plankton (n = 7) | 2 = willow (n = 8):
## U = 28.000, W = 0.000, p = 0.001, Z = -3.240
## effect-size r = 0.837
## rank-mean(1) = 4.00
## rank-mean(2) = 11.50par(mfrow=c(1,2))
boxplot(d15N~Plant, data=topes.controls.non.enrich)
boxplot(C.N~Plant, data=topes.controls.non.enrich)
############# separate plant dfs
#### Sage d15N
topes.controls.sage<-topes.controls[(topes.controls$Plant=="sage"),]
topes.controls.sage$Plant<-droplevels(topes.controls.sage$Plant)
topes.controls.sage$Burn.Trt<-droplevels(topes.controls.sage$Burn.Trt)
# how do different types of sage compare across burn/unburn
# first, no difference between burned or very burned sage
anova(lm(d15N~Burn.Trt, data=topes.controls.sage[!(topes.controls.sage$Burn.Trt=="unburned"),]))## Analysis of Variance Table
##
## Response: d15N
## Df Sum Sq Mean Sq F value Pr(>F)
## Burn.Trt 1 4078 4077.6 1.1836 0.298
## Residuals 12 41339 3444.9# convert to just 2 levels, no difference here either
topes.controls.sage$Burn.Unb<-ifelse(topes.controls.sage$Burn.Trt=="burned", "burned",
ifelse(topes.controls.sage$Burn.Trt=="very burned", "burned",
"unburned"))
mod.sage<-lm(d15N~Burn.Trt, data=topes.controls.sage) # keep at 3 levels
anova(mod.sage) ## Analysis of Variance Table
##
## Response: d15N
## Df Sum Sq Mean Sq F value Pr(>F)
## Burn.Trt 2 5178 2589.1 0.9125 0.4227
## Residuals 15 42561 2837.4# no difference in d15N for burned, unburned, very burned sage (p=0.423)
#### Sage C.N
mod.sage.CN<-lm(C.N~Burn.Trt, data=topes.controls.sage)
anova(mod.sage.CN) ## Analysis of Variance Table
##
## Response: C.N
## Df Sum Sq Mean Sq F value Pr(>F)
## Burn.Trt 2 3010.5 1505.26 11.32 0.001423 **
## Residuals 13 1728.7 132.98
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1posthoc<-emmeans(mod.sage.CN, ~Burn.Trt)
multcomp::cld(posthoc, Letters=letters)## Burn.Trt emmean SE df lower.CL upper.CL .group
## burned 37.5 4.08 13 28.7 46.3 a
## unburned 43.8 5.77 13 31.3 56.2 a
## veryburned 70.7 5.77 13 58.2 83.1 b
##
## Confidence level used: 0.95
## P value adjustment: tukey method for comparing a family of 3 estimates
## significance level used: alpha = 0.05
## NOTE: If two or more means share the same grouping letter,
## then we cannot show them to be different.
## But we also did not show them to be the same.# Sage: difference in unburned, burned, very burned for C:N
par(mfrow=c(1,2))
boxplot(d15N~Burn.Trt, data=topes.controls.sage)
boxplot(C.N~Burn.Trt, data=topes.controls.sage)
###########################
#### Willow d15N
topes.controls.will<-topes.controls[(topes.controls$Plant=="willow"),]
topes.controls.will$Plant<-droplevels(topes.controls.will$Plant)
topes.controls.will$Burn.Trt<-droplevels(topes.controls.will$Burn.Trt)
mod<-lm(C.N~Burn.Trt, data=topes.controls.will)
anova(mod) ## Analysis of Variance Table
##
## Response: C.N
## Df Sum Sq Mean Sq F value Pr(>F)
## Burn.Trt 1 14.773 14.7729 5.2794 0.06129 .
## Residuals 6 16.789 2.7982
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1# no difference in burned/unburned willow d15N
#### Willow C.N
mod<-lm(C.N~Burn.Trt, data=topes.controls.will)
anova(mod) ## Analysis of Variance Table
##
## Response: C.N
## Df Sum Sq Mean Sq F value Pr(>F)
## Burn.Trt 1 14.773 14.7729 5.2794 0.06129 .
## Residuals 6 16.789 2.7982
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1# Willow: no difference in unburned, burned C:N (p=0.061)
par(mfrow=c(1,2))
boxplot(d15N~Burn.Trt, data=topes.controls.will)
boxplot(C.N~Burn.Trt, data=topes.controls.will)
######### make summary dfs
# summarize by plants
d15N.plant<-aggregate(d15N~Plant, topes.controls, FUN=mean)
d15N.plantSD<-aggregate(d15N~Plant, topes.controls, FUN=length)
d15N.plant[3]<-d15N.plantSD[2]
colnames(d15N.plant)<-c("Plant", "d15N", "SD")
CN.plant<-aggregate(C.N~Plant, topes.controls, FUN=mean)
CN.plantSD<-aggregate(C.N~Plant, topes.controls, FUN=sd)
CN.plant[3]<-CN.plantSD[2]
colnames(CN.plant)<-c("Plant", "C.N", "SD")
# summary df d15N
d15N.cont<-aggregate(d15N~Type, topes.controls, FUN=mean)
d15N.cont.n<-aggregate(d15N~Type, topes.controls, FUN=length)
d15N.cont.SD<-aggregate(d15N~Type, topes.controls, FUN=sd)
d15N.cont[3]<- d15N.cont.SD[2]
d15N.cont[4]<- d15N.cont.n[2]
colnames(d15N.cont)<-c("Type", "d15N", "SD", "n")
# summary df control C:N
CN.cont<-aggregate(C.N~Type, topes.controls, FUN=mean)
CN.cont.n<-aggregate(C.N~Type, topes.controls, FUN=length)
CN.cont.SD<-aggregate(C.N~Type, topes.controls, FUN=sd)
CN.cont[3]<- CN.cont.SD[2]
CN.cont[4]<- CN.cont.n[2]
colnames(CN.cont)<-c("Type", "C.N", "SD", "n")make boxplots of control sample d15N and C:N
########## control plots
# set levels
topes.controls$Type<-factor(topes.controls$Type,
levels=c("blank", "plankton.stock",
"willow.unburned", "willow.burned",
"sage.unburned", "sage.burned",
"sage.veryburned", "sage.stem.burned"))
#### controls d15N boxplot
iso.plot.control.d15N<-ggplot(data=topes.controls, aes(x=Type, y=d15N, fill=Type)) +
geom_boxplot() +
geom_point(pch = 21, position = position_jitterdodge(), alpha=0.6)+
ylab(expression(paste(delta^{15}, N, " (\u2030, air)"))) +
scale_fill_manual(values = c("azure2", "cornflowerblue",
"darkgoldenrod1", "indianred2",
"aquamarine3", "antiquewhite3",
"darkolivegreen4", "lightsalmon")) +
xlab("control types") + Fig.formatting +
theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust=1))
###### control C:N
topes.controls.CN<-topes.controls %>% drop_na(C.N) # drop the NAs for C.N, makes plotting problematic
iso.plot.control.CN<-ggplot(data=topes.controls.CN, aes(x=Type, y=C.N, fill=Type)) +
geom_boxplot() +
geom_point(pch = 21, position = position_jitterdodge(), alpha=0.6)+
ylab("C:N") +
scale_fill_manual(values = c("cornflowerblue",
"darkgoldenrod1", "indianred2",
"aquamarine3", "antiquewhite3",
"darkolivegreen4", "lightsalmon")) +
xlab("control types") + Fig.formatting +
theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust=1))
####### combine plots
# legend
extract.legend.cont <- get_legend(
# create some space to the left of the legend
iso.plot.control.d15N + theme(legend.box.margin = margin(0, 0, 0, 10)))
## combine
control.iso.alltime<-
plot_grid(iso.plot.control.d15N + theme(legend.position = "none"),
iso.plot.control.CN + theme(legend.position = "none"),
extract.legend.cont, rel_widths = c(8,8,3), ncol=3)
ggsave("figures/iso.controls.pdf", encod="MacRoman", height=5, width=10)Use a mixing model to calculate % sage for plankton
# mixing model
head(topes.trt)
topes.trt<-droplevels(topes.trt)
### values for controls
d15N.cont # summary mean d15N by all controls
d15N.plant # summary by plants, # stock plantkon 11, # sage 296, # willow 13
# summary mean atom percent enrichment
F.cont<-aggregate(at.P..15N ~ Type, topes.controls, mean)
F.plant<-aggregate(at.P..15N~Plant, topes.controls, FUN=mean)
# sage ~0.475
# willow 0.371
# 2 source mixing model (Post 2002), used d15N values here
# alpha = percent Sage from food web 1
# %Sage = (d15N sample - d15N base 2 [i.e., no-label food])/ (d15N sage food 1 - source 2)
# d15N values of base 2 = 11 permil for algae/plankton stock/willow
# d15N value of base 1 = 298 permil for sage
# framed differently from Robinson 2001, TREE
# xtracer = frction of tracer
# Xtracer = (d15N-sample - d15N background) / (d15N-tracer - d15N-background)
topes.trt$percent.sage<-(topes.trt$d15N-12)/(296-12)*100
# unicode text for micrometer = \u03BC, use this in legendmodel for % sage
############# all plankton T1
m1.T1.sage <- gam(percent.sage ~ Treatment + Type +
s(plant.mass..g, by=Treatment),
subset = Time.point=="T1", data=topes.trt, method="REML", family="gaussian")
m2.T1.sage <- gam(percent.sage ~ Treatment + Type +
s(plant.mass..g),
subset = Time.point=="T1", data=topes.trt, method="REML", family="gaussian")
m3.T1.sage <- gam(percent.sage ~
s(plant.mass..g),
subset = Time.point=="T1", data=topes.trt, method="REML", family="gaussian")
AIC.sage.T1<-AIC(m1.T1.sage, m2.T1.sage, m3.T1.sage)
## additive model best fit
summary(m1.T1.sage)
anova.gam(m1.T1.sage)
gam.check(m1.T1.sage, rep=1000)draw(m1.T1.sage)concrvity(m1.T1.sage)
par(mfrow = c(1, 2))
plot(m1.T1.sage, all.terms = TRUE, page=1)# model predictions
per.Sage.diff.T1<-plot_difference(
m1.T1.sage,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
# model for smoothing
msmooth.T1.sage<- gam(percent.sage ~ Treatment +
s(plant.mass..g, by=Treatment),
subset = Time.point=="T1", data=topes.trt, method="REML", family="gaussian")
#plot for the model output on rawdata
per.Sage.T1.mod.plot<-
plot_smooths(
model = msmooth.T1.sage,
series = plant.mass..g,
comparison = Treatment
) + theme(legend.position = "none") +
geom_point(data=topes.trt[(topes.trt$Time.point=="T1"),],
aes(x=plant.mass..g, y=percent.sage, color=Treatment, shape=Type)) +
scale_shape_manual(name="Plankton", values = c(17, 16),
labels = c(expression(paste("> 63"~mu,"m")),
expression(paste("< 63"~mu,"m")))) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
ylab("% Sage")+
xlab("plant material (g)") +
ggtitle("Time-1") +
coord_cartesian(ylim=c(0, 100)) +
Fig.formatting +
theme(legend.key.size = unit(1,"line"))
# overall an effect of burning with both smoothers being significant by treatment
# no effect of type, POM and plankton with similar d15N values
############# all plankton T2
m1.T2.sage <- gam(percent.sage ~ Treatment + Type +
s(plant.mass..g, by=Treatment),
subset = Time.point=="T2", data=topes.trt, method="REML", family="gaussian")
m2.T2.sage <- gam(percent.sage ~ Treatment + Type +
s(plant.mass..g),
subset = Time.point=="T2", data=topes.trt, method="REML", family="gaussian")
m3.T2.sage <- gam(percent.sage ~
s(plant.mass..g),
subset = Time.point=="T2", data=topes.trt, method="REML", family="gaussian")
AIC.sage.T2<-AIC(m1.T2.sage, m2.T2.sage, m3.T2.sage)
# m2.T2.sage preferred
# compare across
summary(m1.T2.sage)
anova.gam(m1.T2.sage)
gam.check(m1.T2.sage, rep=1000)draw(m1.T2.sage)concrvity(m1.T2.sage)
par(mfrow = c(1, 2))
plot(m1.T2.sage, all.terms = TRUE, page=1)# model predictions
per.Sage.diff.T2<-plot_difference(
m1.T2.sage,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
# model for smoothing
msmooth.T2.sage<- gam(percent.sage ~ Treatment +
s(plant.mass..g, by=Treatment),
subset = Time.point=="T2", data=topes.trt, method="REML", family="gaussian")
#plot for the model output on rawdata
per.Sage.T2.mod.plot<-
plot_smooths(
model = msmooth.T2.sage,
series = plant.mass..g,
comparison = Treatment
) + theme(legend.position = "none") +
geom_point(data=topes.trt[(topes.trt$Time.point=="T2"),],
aes(x=plant.mass..g, y=percent.sage, color=Treatment, shape=Type)) +
scale_shape_manual(name="Plankton", values = c(17, 16),
labels = c(expression(paste("> 63"~mu,"m")),
expression(paste("< 63"~mu,"m")))) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
ylab("% Sage")+
xlab("plant material (g)") +
ggtitle("Time-2") +
coord_cartesian(ylim=c(0, 100)) +
Fig.formatting +
theme(legend.key.size = unit(1,"line"))
# effect of treatment (p<0.001) but not Type (p=0.321)
# smoother significant for both termsmodel plots for percent sage mixing model
## AIC table
mod.sag.top<-rep(c( "Treatment + Type + s(plant.mass..g, by=Treatment)",
"Treatment + Type + s(plant.mass..g)",
"s(plant.mass..g)"), times=2)
mod.sag.df<- data.frame(mod.sag.top)
AIC.sag.topes<-bind_rows(AIC.sage.T1, AIC.sage.T2)
AIC.sag.mod<-cbind(mod.sag.df, AIC.sag.topes)
write.csv(AIC.sag.mod, "output/AIC models/AIC.sag.mod.csv")
# legend
extract.legend.mix <- get_legend(
# create some space to the left of the legend
per.Sage.T2.mod.plot + theme(legend.box.margin = margin(0, 0, 0, 10)))
## combine
sage.mix.model<- plot_grid(per.Sage.T1.mod.plot + theme(legend.position = "none"),
per.Sage.T2.mod.plot + theme(legend.position = "none"),
extract.legend.mix, rel_widths = c(8,8,3), ncol=3)
ggsave("figures/Isotope.mixmodel.pdf", encod="MacRoman", height=4, width=8)
# and plot difference
sage.mix.model.diff<- plot_grid(per.Sage.diff.T1 + theme(legend.position = "none"),
per.Sage.diff.T2 + theme(legend.position = "none"),
extract.legend.mix, rel_widths = c(8,8,3), ncol=3)
ggsave("figures/Isotope.mix.plotdiff.pdf", encod="MacRoman", height=4, width=8)make d15N plots as well – these follow the % sage, but are informative with the control plots to see the d15N of plankton and the 2 end members.
#### d15N isotope plots for treatments
############# all plankton T1
m1.T1.d15N <- gam(d15N ~ Treatment + Type +
s(plant.mass..g, by=Treatment),
subset = Time.point=="T1", data=topes.trt, method="REML", family="gaussian")
m2.T1.d15N <- gam(d15N ~ Treatment + Type +
s(plant.mass..g),
subset = Time.point=="T1", data=topes.trt, method="REML", family="gaussian")
m3.T1.d15N <- gam(d15N ~
s(plant.mass..g),
subset = Time.point=="T1", data=topes.trt, method="REML", family="gaussian")
AIC.d15N.T1<-AIC(m1.T1.d15N, m2.T1.d15N, m3.T1.d15N)
## additive model best fit
summary(m1.T1.d15N)##
## Family: gaussian
## Link function: identity
##
## Formula:
## d15N ~ Treatment + Type + s(plant.mass..g, by = Treatment)
##
## Parametric coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 96.337 2.610 36.908 < 2e-16 ***
## Treatmentunburned 16.852 3.014 5.591 9.64e-07 ***
## TypePOM 3.954 3.014 1.312 0.196
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 3.560 4.338 136.8 <2e-16 ***
## s(plant.mass..g):Treatmentunburned 3.921 4.762 173.6 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## R-sq.(adj) = 0.961 Deviance explained = 96.7%
## -REML = 227.82 Scale est. = 136.26 n = 60anova.gam(m1.T1.d15N)##
## Family: gaussian
## Link function: identity
##
## Formula:
## d15N ~ Treatment + Type + s(plant.mass..g, by = Treatment)
##
## Parametric Terms:
## df F p-value
## Treatment 1 31.261 9.64e-07
## Type 1 1.721 0.196
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 3.560 4.338 136.8 <2e-16
## s(plant.mass..g):Treatmentunburned 3.921 4.762 173.6 <2e-16gam.check(m1.T1.d15N, rep=1000)
##
## Method: REML Optimizer: outer newton
## full convergence after 5 iterations.
## Gradient range [-2.454513e-09,1.329425e-10]
## (score 227.823 & scale 136.2639).
## Hessian positive definite, eigenvalue range [0.5347593,27.64001].
## Model rank = 21 / 21
##
## Basis dimension (k) checking results. Low p-value (k-index<1) may
## indicate that k is too low, especially if edf is close to k'.
##
## k' edf k-index p-value
## s(plant.mass..g):Treatmentburned 9.00 3.56 1 0.47
## s(plant.mass..g):Treatmentunburned 9.00 3.92 1 0.38draw(m1.T1.d15N)
concrvity(m1.T1.d15N)## # A tibble: 9 × 3
## type term concurvity
## <chr> <chr> <dbl>
## 1 worst para 6.67e- 1
## 2 worst s(plant.mass..g):Treatmentburned 4.24e-26
## 3 worst s(plant.mass..g):Treatmentunburned 4.28e-26
## 4 observed para 6.67e- 1
## 5 observed s(plant.mass..g):Treatmentburned 7.44e-31
## 6 observed s(plant.mass..g):Treatmentunburned 3.91e-30
## 7 estimate para 6.67e- 1
## 8 estimate s(plant.mass..g):Treatmentburned 1.27e-28
## 9 estimate s(plant.mass..g):Treatmentunburned 1.26e-28par(mfrow = c(1, 2))
plot(m1.T1.d15N, all.terms = TRUE, page=1)
# model predictions
d15N.diff.T1<-plot_difference(
m1.T1.d15N,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
# model for smoothing
msmooth.T1.d15N<- gam(d15N ~ Treatment +
s(plant.mass..g, by=Treatment),
subset = Time.point=="T1", data=topes.trt, method="REML", family="gaussian")
#plot for the model output on rawdata
d15N.T1.mod.plot<-
plot_smooths(
model = msmooth.T1.d15N,
series = plant.mass..g,
comparison = Treatment
) + theme(legend.position = "none") +
geom_point(data=topes.trt[(topes.trt$Time.point=="T1"),],
aes(x=plant.mass..g, y=d15N, color=Treatment, shape=Type)) +
scale_shape_manual(name="Plankton", values = c(17, 16),
labels = c(expression(paste("> 63"~mu,"m")),
expression(paste("< 63"~mu,"m")))) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
ylab(expression(paste(delta^{15}, N, " (\u2030, air)"))) +
xlab("plant material (g)") +
ggtitle("Time-1") +
coord_cartesian(ylim=c(0, 250)) +
Fig.formatting +
theme(legend.key.size = unit(1,"line"))
# overall an effect of burning with both smoothers being significant by treatment
# no effect of type, POM and plankton with similar d15N values
############# all plankton T2
m1.T2.d15N <- gam(d15N ~ Treatment + Type +
s(plant.mass..g, by=Treatment),
subset = Time.point=="T2", data=topes.trt, method="REML", family="gaussian")
m2.T2.d15N <- gam(d15N ~ Treatment + Type +
s(plant.mass..g),
subset = Time.point=="T2", data=topes.trt, method="REML", family="gaussian")
m3.T2.d15N <- gam(d15N ~
s(plant.mass..g),
subset = Time.point=="T2", data=topes.trt, method="REML", family="gaussian")
AIC.d15N.T2<-AIC(m1.T2.d15N, m2.T2.d15N, m3.T2.d15N)
## additive model best fit
summary(m1.T2.d15N)##
## Family: gaussian
## Link function: identity
##
## Formula:
## d15N ~ Treatment + Type + s(plant.mass..g, by = Treatment)
##
## Parametric coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 91.147 2.944 30.960 < 2e-16 ***
## Treatmentunburned 12.290 3.398 3.617 0.000704 ***
## TypePOM -3.409 3.402 -1.002 0.321237
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 4.669 5.621 79.45 <2e-16 ***
## s(plant.mass..g):Treatmentunburned 3.340 4.082 125.45 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## R-sq.(adj) = 0.943 Deviance explained = 95.3%
## -REML = 235.35 Scale est. = 172.72 n = 60anova.gam(m1.T2.d15N)##
## Family: gaussian
## Link function: identity
##
## Formula:
## d15N ~ Treatment + Type + s(plant.mass..g, by = Treatment)
##
## Parametric Terms:
## df F p-value
## Treatment 1 13.082 0.000704
## Type 1 1.004 0.321237
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 4.669 5.621 79.45 <2e-16
## s(plant.mass..g):Treatmentunburned 3.340 4.082 125.45 <2e-16gam.check(m1.T2.d15N, rep=1000)
##
## Method: REML Optimizer: outer newton
## full convergence after 4 iterations.
## Gradient range [-7.936647e-08,5.277139e-09]
## (score 235.3485 & scale 172.7225).
## Hessian positive definite, eigenvalue range [1.026571,27.67886].
## Model rank = 21 / 21
##
## Basis dimension (k) checking results. Low p-value (k-index<1) may
## indicate that k is too low, especially if edf is close to k'.
##
## k' edf k-index p-value
## s(plant.mass..g):Treatmentburned 9.00 4.67 0.95 0.32
## s(plant.mass..g):Treatmentunburned 9.00 3.34 0.95 0.28draw(m1.T2.d15N)
concrvity(m1.T2.d15N)## # A tibble: 9 × 3
## type term concurvity
## <chr> <chr> <dbl>
## 1 worst para 0.669
## 2 worst s(plant.mass..g):Treatmentburned 0.00464
## 3 worst s(plant.mass..g):Treatmentunburned 0.0148
## 4 observed para 0.669
## 5 observed s(plant.mass..g):Treatmentburned 0.00100
## 6 observed s(plant.mass..g):Treatmentunburned 0.00307
## 7 estimate para 0.669
## 8 estimate s(plant.mass..g):Treatmentburned 0.000543
## 9 estimate s(plant.mass..g):Treatmentunburned 0.00169par(mfrow = c(1, 2))
plot(m1.T2.d15N, all.terms = TRUE, page=1)
# model predictions
d15N.diff.T2<-plot_difference(
m1.T2.d15N,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
# model for smoothing
msmooth.T2.d15N<- gam(d15N ~ Treatment +
s(plant.mass..g, by=Treatment),
subset = Time.point=="T2", data=topes.trt, method="REML", family="gaussian")
#plot for the model output on rawdata
d15N.T2.mod.plot<-
plot_smooths(
model = msmooth.T2.d15N,
series = plant.mass..g,
comparison = Treatment
) + theme(legend.position = "none") +
geom_point(data=topes.trt[(topes.trt$Time.point=="T2"),],
aes(x=plant.mass..g, y=d15N, color=Treatment, shape=Type)) +
scale_shape_manual(name="Plankton", values = c(17, 16),
labels = c(expression(paste("> 63"~mu,"m")),
expression(paste("< 63"~mu,"m")))) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
ylab(expression(paste(delta^{15}, N, " (\u2030, air)"))) +
xlab("plant material (g)") +
ggtitle("Time-2") +
coord_cartesian(ylim=c(0, 250)) +
Fig.formatting +
theme(legend.key.size = unit(1,"line"))
d15N.model<- plot_grid(d15N.T1.mod.plot + theme(legend.position = "none"),
d15N.T2.mod.plot + theme(legend.position = "none"),
extract.legend.mix, rel_widths = c(8,8,3), ncol=3)
ggsave("figures/d15N.model.pdf", encod="MacRoman", height=5, width=10)
################
mod.d15N<-rep(c("Treatment + Type + s(plant.mass..g, by=Treatment)",
"Treatment + Type + s(plant.mass..g)",
"s(plant.mass..g)"), times=2)
mod.d15N.df<- data.frame(mod.d15N)
AIC.d15N<-bind_rows(AIC.d15N.T1, AIC.d15N.T2)
AIC.d15N.mod<-cbind(mod.d15N.df, AIC.d15N)
write.csv(AIC.d15N.mod, "output/AIC models/AIC.d15N.mod.csv")#################
###
POM<-topes.trt[(topes.trt$Type=="POM"),]
plankton<-topes.trt[(topes.trt$Type=="plankton"),]
topes <- POM %>% inR_join( plankton,
by=c('Time.point','Treatment', 'plant.mass..g', 'Tank'))
names(topes)[names(topes) == 'percent.sage.x'] <- 'POM.per.sage'
names(topes)[names(topes) == 'percent.sage.y'] <- 'plank.per.sage'
topes<- topes %>%
select(Time.point, Treatment, plant.mass..g, Tank, POM.per.sage, plank.per.sage)
topes$TTE.diff<- topes$POM.per.sage - topes$plank.per.sage
topes$TTE.per <- (topes$plank.per.sage / topes$POM.per.sage)*100
topes$TTE.per[topes$TTE.per < 0] <-"" # remove outlier
topes$TTE.per<-as.numeric(topes$TTE.per)
# and again...
topes$TTE.per[topes$TTE.per > 200] <-"" # remove outlier
topes$TTE.per<-as.numeric(topes$TTE.per)
################### TTE as difference
## run some models
m1.T1.TTE <- gam(TTE.diff ~ Treatment +
s(plant.mass..g, by=Treatment),
subset = Time.point=="T1", data=topes, method="REML", family="gaussian")
m2.T1.TTE <- gam(TTE.diff ~ Treatment +
s(plant.mass..g),
subset = Time.point=="T1", data=topes, method="REML", family="gaussian")
m3.T1.TTE <- gam(TTE.diff ~
s(plant.mass..g),
subset = Time.point=="T1", data=topes, method="REML", family="gaussian")
AIC.TTE.T1<-AIC(m1.T1.TTE, m2.T1.TTE, m3.T1.TTE)
## first model best fit
summary(m1.T1.TTE)
anova.gam(m1.T1.TTE)
gam.check(m1.T1.TTE, rep=1000)
draw(m1.T1.TTE)
concrvity(m1.T1.TTE)
par(mfrow = c(1, 2))
plot(m1.T1.TTE, all.terms = TRUE, page=1)
# model predictions
TTE.T1<-plot_difference(
m1.T1.TTE,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
# model for smoothing
msmooth.T1.TTE<- gam(TTE.diff ~ Treatment +
s(plant.mass..g, by=Treatment),
subset = Time.point=="T1", data=topes, method="REML", family="gaussian")
#plot for the model output on rawdata
TTE.T1.mod.plot<-
plot_smooths(
model = msmooth.T1.TTE,
series = plant.mass..g,
comparison = Treatment
) + theme(legend.position = "none") +
geom_point(data=topes[(topes$Time.point=="T1"),],
aes(x=plant.mass..g, y=TTE.diff, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
ylab("TTE= POM %sage - Plank %sage")+
xlab("plant material (g)") +
ggtitle("Time-1") +
coord_cartesian(ylim=c(-10, 25)) +
Fig.formatting +
theme(legend.key.size = unit(1,"line"))
################### TTE as difference TIME 2
## run some models
m1.T2.TTE <- gam(TTE.diff ~ Treatment +
s(plant.mass..g, by=Treatment),
subset = Time.point=="T2", data=topes, method="REML", family="gaussian")
m2.T2.TTE <- gam(TTE.diff ~ Treatment +
s(plant.mass..g),
subset = Time.point=="T2", data=topes, method="REML", family="gaussian")
m3.T2.TTE <- gam(TTE.diff ~
s(plant.mass..g),
subset = Time.point=="T2", data=topes, method="REML", family="gaussian")
AIC.TTE.T2<-AIC(m1.T2.TTE, m2.T2.TTE, m3.T2.TTE)
## first model best fit
summary(m1.T2.TTE)
anova.gam(m1.T2.TTE)
gam.check(m1.T2.TTE, rep=1000)
draw(m1.T2.TTE)
concrvity(m1.T2.TTE)
par(mfrow = c(1, 2))
plot(m1.T2.TTE, all.terms = TRUE, page=1)
# model predictions
TTE.T2<-plot_difference(
m1.T2.TTE,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
# model for smoothing
msmooth.T2.TTE<- gam(TTE.diff ~ Treatment +
s(plant.mass..g, by=Treatment),
subset = Time.point=="T2", data=topes, method="REML", family="gaussian")
#plot for the model output on rawdata
TTE.T2.mod.plot<-
plot_smooths(
model = msmooth.T2.TTE,
series = plant.mass..g,
comparison = Treatment
) + theme(legend.position = "none") +
geom_point(data=topes[(topes$Time.point=="T2"),],
aes(x=plant.mass..g, y=TTE.diff, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
ylab("TTE= POM %sage - Plank %sage")+
xlab("plant material (g)") +
ggtitle("Time-2") +
coord_cartesian(ylim=c(-10, 25)) +
Fig.formatting +
theme(legend.key.size = unit(1,"line"))
############ as percent
m1.T1.TTE.per <- gam(TTE.per ~ Treatment +
s(plant.mass..g, by=Treatment),
subset = Time.point=="T1", data=topes, method="REML", family="gaussian")
m2.T1.TTE.per <- gam(TTE.per ~ Treatment +
s(plant.mass..g),
subset = Time.point=="T1", data=topes, method="REML", family="gaussian")
m3.T1.TTE.per <- gam(TTE.per ~
s(plant.mass..g),
subset = Time.point=="T1", data=topes, method="REML", family="gaussian")
AIC.TTE.per.T1<-AIC(m1.T1.TTE.per, m2.T1.TTE.per, m3.T1.TTE.per)
## first model best fit
summary(m1.T1.TTE.per)
anova.gam(m1.T1.TTE.per)
gam.check(m1.T1.TTE.per, rep=1000)
draw(m1.T1.TTE.per)
concrvity(m1.T1.TTE.per)
par(mfrow = c(1, 2))
plot(m1.T1.TTE.per, all.terms = TRUE, page=1)
# model predictions
TTE.per.T1<-plot_difference(
m1.T1.TTE.per,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
# model for smoothing
msmooth.T1.TTE.per<- gam(TTE.per ~ Treatment +
s(plant.mass..g, by=Treatment),
subset = Time.point=="T1", data=topes, method="REML", family="gaussian")
#plot for the model output on rawdata
TTE.per.T1.mod.plot<-
plot_smooths(
model = msmooth.T1.TTE.per,
series = plant.mass..g,
comparison = Treatment
) + theme(legend.position = "none") +
geom_point(data=topes[(topes$Time.point=="T1"),],
aes(x=plant.mass..g, y=TTE.per, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
ylab("TTE= (plank - POM) *100")+
xlab("plant material (g)") +
ggtitle("Time-1") +
coord_cartesian(ylim=c(0, 200)) +
Fig.formatting +
theme(legend.key.size = unit(1,"line"))
### Time 2
m1.T2.TTE.per <- gam(TTE.per ~ Treatment +
s(plant.mass..g, by=Treatment),
subset = Time.point=="T2", data=topes, method="REML", family="gaussian")
m2.T2.TTE.per <- gam(TTE.per ~ Treatment +
s(plant.mass..g),
subset = Time.point=="T2", data=topes, method="REML", family="gaussian")
m3.T2.TTE.per <- gam(TTE.per ~
s(plant.mass..g),
subset = Time.point=="T2", data=topes, method="REML", family="gaussian")
AIC.TTE.per.T2<-AIC(m1.T2.TTE.per, m2.T2.TTE.per, m3.T2.TTE.per)
## first model best fit
summary(m1.T2.TTE.per)
anova.gam(m1.T2.TTE.per)
gam.check(m1.T2.TTE.per, rep=1000)
draw(m1.T2.TTE.per)
concrvity(m1.T2.TTE.per)
par(mfrow = c(1, 2))
plot(m1.T2.TTE.per, all.terms = TRUE, page=1)
# model predictions
TTE.per.T2<-plot_difference(
m1.T2.TTE.per,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
# model for smoothing
msmooth.T2.TTE.per<- gam(TTE.per ~ Treatment +
s(plant.mass..g, by=Treatment),
subset = Time.point=="T2", data=topes, method="REML", family="gaussian")
#plot for the model output on rawdata
TTE.per.T2.mod.plot<-
plot_smooths(
model = msmooth.T2.TTE.per,
series = plant.mass..g,
comparison = Treatment
) + theme(legend.position = "none") +
geom_point(data=topes[(topes$Time.point=="T2"),],
aes(x=plant.mass..g, y=TTE.per, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
ylab("TTE= (plank - POM) *100")+
xlab("plant material (g)") +
ggtitle("Time-2") +
coord_cartesian(ylim=c(0, 200)) +
Fig.formatting +
theme(legend.key.size = unit(1,"line"))
## combine
TTE.diff.plot<- plot_grid(TTE.T1.mod.plot + theme(legend.position = "none"),
TTE.T2.mod.plot + theme(legend.position = "none"),
extract.legend, rel_widths = c(8,8,3), ncol=3)
TTE.per.plot<- plot_grid(TTE.per.T1.mod.plot + theme(legend.position = "none"),
TTE.per.T2.mod.plot + theme(legend.position = "none"),
extract.legend, rel_widths = c(8,8,3), ncol=3)
ggsave("figures/TTE.diff.plot.pdf", encod="MacRoman", height=4, width=8)
ggsave("figures/TTE.per.plot.pdf", encod="MacRoman", height=4, width=8)Plant material
Plant material for the starting material (sage or willow, stems or sticks). This is useful in determining how fire impacted the nutrients in the plant material.
First, run some stats to see what is happening and where differences lie.
plant.nut<-read.csv("data/Pyro_plant material_elemental.csv")
fac<-c("Sample.Name", "type", "plant", "treatment", "rep") # make factors
plant.nut[fac]<-lapply(plant.nut[fac],factor)
sage.nut<-plant.nut[(plant.nut$plant=="sage"),]
will.nut<-plant.nut[(plant.nut$plant=="willow"),]
### Sage ###
######### %N
plant.N.sag<-lm(N~treatment*type, data=sage.nut)
Anova(plant.N.sag, type=3) # 2 way interaction and main effects## Anova Table (Type III tests)
##
## Response: N
## Sum Sq Df F value Pr(>F)
## (Intercept) 7.0227 1 2171.5213 4.971e-11 ***
## treatment 0.0280 1 8.6632 0.018615 *
## type 0.2604 1 80.5246 1.894e-05 ***
## treatment:type 0.0705 1 21.8099 0.001602 **
## Residuals 0.0259 8
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1posthoc<-emmeans(plant.N.sag, ~treatment| type)
multcomp::cld(posthoc, Letters=letters)## type = leaf:
## treatment emmean SE df lower.CL upper.CL .group
## burned 1.530 0.0328 8 1.454 1.61 a
## unburned 1.667 0.0328 8 1.591 1.74 b
##
## type = stem:
## treatment emmean SE df lower.CL upper.CL .group
## unburned 0.943 0.0328 8 0.868 1.02 a
## burned 1.113 0.0328 8 1.038 1.19 b
##
## Confidence level used: 0.95
## significance level used: alpha = 0.05
## NOTE: If two or more means share the same grouping letter,
## then we cannot show them to be different.
## But we also did not show them to be the same.########## %P
plant.P.sag<-lm(P~treatment*type, data=sage.nut)
Anova(plant.P.sag, type=3) # just type## Anova Table (Type III tests)
##
## Response: P
## Sum Sq Df F value Pr(>F)
## (Intercept) 0.314928 1 804.0715 2.586e-09 ***
## treatment 0.000771 1 1.9677 0.198292
## type 0.007561 1 19.3060 0.002306 **
## treatment:type 0.000056 1 0.1438 0.714371
## Residuals 0.003133 8
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1########## %K
plant.K.sag<-lm(K~treatment*type, data=sage.nut)
Anova(plant.K.sag, type=3) # type and treatment## Anova Table (Type III tests)
##
## Response: K
## Sum Sq Df F value Pr(>F)
## (Intercept) 12.5256 1 372.2328 5.404e-08 ***
## treatment 0.2563 1 7.6157 0.02469 *
## type 0.1803 1 5.3571 0.04934 *
## treatment:type 0.0736 1 2.1882 0.17733
## Residuals 0.2692 8
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1posthoc<-emmeans(plant.K.sag, ~treatment)
multcomp::cld(posthoc, Letters=letters)## treatment emmean SE df lower.CL upper.CL .group
## unburned 1.61 0.0749 8 1.44 1.79 a
## burned 1.87 0.0749 8 1.70 2.04 b
##
## Results are averaged over the levels of: type
## Confidence level used: 0.95
## significance level used: alpha = 0.05
## NOTE: If two or more means share the same grouping letter,
## then we cannot show them to be different.
## But we also did not show them to be the same.########## %S
plant.S.sag<-lm(S~treatment*type, data=sage.nut)
Anova(plant.S.sag, type=3) # type effect## Anova Table (Type III tests)
##
## Response: S
## Sum Sq Df F value Pr(>F)
## (Intercept) 0.67783 1 859.3665 1.986e-09 ***
## treatment 0.00010 1 0.1321 0.7257
## type 0.11152 1 141.3891 2.299e-06 ***
## treatment:type 0.00039 1 0.4885 0.5044
## Residuals 0.00631 8
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1############## Zn ppm
plant.ZN.sag<-lm(ZN~treatment*type, data=sage.nut)
Anova(plant.ZN.sag, type=3) #no effect## Anova Table (Type III tests)
##
## Response: ZN
## Sum Sq Df F value Pr(>F)
## (Intercept) 19959.4 1 15.1723 0.004576 **
## treatment 3073.6 1 2.3364 0.164903
## type 3280.7 1 2.4938 0.152947
## treatment:type 1236.3 1 0.9398 0.360731
## Residuals 10524.1 8
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1posthoc<-emmeans(plant.ZN.sag, ~treatment| type)
multcomp::cld(posthoc, Letters=letters)## type = leaf:
## treatment emmean SE df lower.CL upper.CL .group
## unburned 36.3 20.9 8 -12.0 84.6 a
## burned 81.6 20.9 8 33.3 129.9 a
##
## type = stem:
## treatment emmean SE df lower.CL upper.CL .group
## unburned 30.1 20.9 8 -18.2 78.4 a
## burned 34.8 20.9 8 -13.5 83.1 a
##
## Confidence level used: 0.95
## significance level used: alpha = 0.05
## NOTE: If two or more means share the same grouping letter,
## then we cannot show them to be different.
## But we also did not show them to be the same.###############################################
###############################################
### Willow ###
######### %N
plant.N.wil<-lm(N~treatment*type, data=will.nut)
Anova(plant.N.wil, type=3) # main effects## Anova Table (Type III tests)
##
## Response: N
## Sum Sq Df F value Pr(>F)
## (Intercept) 8.7381 1 644.0658 3.766e-08 ***
## treatment 0.0817 1 6.0194 0.04388 *
## type 1.4538 1 107.1530 1.703e-05 ***
## treatment:type 0.0436 1 3.2119 0.11620
## Residuals 0.0950 7
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1########## %P
plant.P.wil<-lm(P~treatment*type, data=will.nut)
Anova(plant.P.wil, type=3) # type and treatment## Anova Table (Type III tests)
##
## Response: P
## Sum Sq Df F value Pr(>F)
## (Intercept) 0.156865 1 447.9429 1.323e-07 ***
## treatment 0.006403 1 18.2834 0.003674 **
## type 0.007239 1 20.6703 0.002647 **
## treatment:type 0.000880 1 2.5131 0.156921
## Residuals 0.002451 7
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1posthoc<-emmeans(plant.P.wil, ~treatment)
multcomp::cld(posthoc, Letters=letters)## treatment emmean SE df lower.CL upper.CL .group
## unburned 0.143 0.00764 7 0.125 0.161 a
## burned 0.190 0.00854 7 0.170 0.210 b
##
## Results are averaged over the levels of: type
## Confidence level used: 0.95
## significance level used: alpha = 0.05
## NOTE: If two or more means share the same grouping letter,
## then we cannot show them to be different.
## But we also did not show them to be the same.########## %K
plant.K.wil<-lm(K~treatment*type, data=will.nut)
Anova(plant.K.wil, type=3) # type and treatment## Anova Table (Type III tests)
##
## Response: K
## Sum Sq Df F value Pr(>F)
## (Intercept) 4.6128 1 548.0120 6.59e-08 ***
## treatment 0.0676 1 8.0344 0.0252428 *
## type 0.2640 1 31.3583 0.0008161 ***
## treatment:type 0.0015 1 0.1725 0.6903484
## Residuals 0.0589 7
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1posthoc<-emmeans(plant.K.wil, ~treatment)
multcomp::cld(posthoc, Letters=letters)## treatment emmean SE df lower.CL upper.CL .group
## unburned 0.817 0.0375 7 0.728 0.905 a
## burned 1.006 0.0419 7 0.906 1.105 b
##
## Results are averaged over the levels of: type
## Confidence level used: 0.95
## significance level used: alpha = 0.05
## NOTE: If two or more means share the same grouping letter,
## then we cannot show them to be different.
## But we also did not show them to be the same.########## %S
plant.S.wil<-lm(S~treatment*type, data=will.nut)
Anova(plant.S.wil, type=3) # main and interactions## Anova Table (Type III tests)
##
## Response: S
## Sum Sq Df F value Pr(>F)
## (Intercept) 0.197633 1 1748.117 1.168e-09 ***
## treatment 0.002521 1 22.303 0.002151 **
## type 0.042375 1 374.819 2.446e-07 ***
## treatment:type 0.001355 1 11.985 0.010519 *
## Residuals 0.000791 7
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1posthoc<-emmeans(plant.S.wil, ~treatment| type)
multcomp::cld(posthoc, Letters=letters)## type = leaf:
## treatment emmean SE df lower.CL upper.CL .group
## unburned 0.2157 0.00614 7 0.2012 0.2302 a
## burned 0.2567 0.00614 7 0.2422 0.2712 b
##
## type = stem:
## treatment emmean SE df lower.CL upper.CL .group
## burned 0.0688 0.00752 7 0.0510 0.0865 a
## unburned 0.0728 0.00614 7 0.0583 0.0873 a
##
## Confidence level used: 0.95
## significance level used: alpha = 0.05
## NOTE: If two or more means share the same grouping letter,
## then we cannot show them to be different.
## But we also did not show them to be the same.############## Zn ppm
plant.ZN.wil<-lm(ZN~treatment*type, data=will.nut)
Anova(plant.ZN.wil, type=3) #no effect## Anova Table (Type III tests)
##
## Response: ZN
## Sum Sq Df F value Pr(>F)
## (Intercept) 124440 1 196.4564 2.229e-06 ***
## treatment 22363 1 35.3043 0.0005748 ***
## type 21956 1 34.6631 0.0006071 ***
## treatment:type 5340 1 8.4306 0.0228697 *
## Residuals 4434 7
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1posthoc<-emmeans(plant.ZN.wil, ~treatment| type)
multcomp::cld(posthoc, Letters=letters)## type = leaf:
## treatment emmean SE df lower.CL upper.CL .group
## unburned 81.6 14.5 7 47.21 115.9 a
## burned 203.7 14.5 7 169.31 238.0 b
##
## type = stem:
## treatment emmean SE df lower.CL upper.CL .group
## unburned 35.8 14.5 7 1.44 70.2 a
## burned 68.4 17.8 7 26.32 110.5 a
##
## Confidence level used: 0.95
## significance level used: alpha = 0.05
## NOTE: If two or more means share the same grouping letter,
## then we cannot show them to be different.
## But we also did not show them to be the same.Make a figure of the elemental analysis data.
plant.nut$int.fac<-factor(interaction(plant.nut$plant, plant.nut$type),
levels=c("sage.leaf", "sage.stem", "willow.leaf", "willow.stem"))
####### figures
N.plot<- ggplot(plant.nut, aes(x=int.fac, y=N, fill=treatment)) +
geom_boxplot(alpha=0.7) +
scale_fill_manual(values = c("brown1", "mediumseagreen")) +
geom_dotplot(binaxis='y', stackdir='center', alpha=0.5, dotsize=0.5,
position=position_dodge(0.75))+
xlab("Plant:Tissue")+
ylab("%N")+
theme_classic() +
theme(axis.text.x = element_text(angle = 45, vjust = 1, hjust=1))
P.plot<- ggplot(plant.nut, aes(x=int.fac, y=P, fill=treatment)) +
geom_boxplot(alpha=0.7) +
scale_fill_manual(values = c("brown1", "mediumseagreen")) +
geom_dotplot(binaxis='y', stackdir='center', alpha=0.5, dotsize=0.5,
position=position_dodge(0.75))+
xlab("Plant:Tissue")+
ylab("%P")+
theme_classic() +
theme(axis.text.x = element_text(angle = 45, vjust = 1, hjust=1))
K.plot<- ggplot(plant.nut, aes(x=int.fac, y=K, fill=treatment)) +
geom_boxplot(alpha=0.7) +
scale_fill_manual(values = c("brown1", "mediumseagreen")) +
geom_dotplot(binaxis='y', stackdir='center', alpha=0.5, dotsize=0.5,
position=position_dodge(0.75))+
xlab("Plant:Tissue")+
ylab("%K")+
theme_classic() +
theme(axis.text.x = element_text(angle = 45, vjust = 1, hjust=1))
S.plot<- ggplot(plant.nut, aes(x=int.fac, y=S, fill=treatment)) +
geom_boxplot(alpha=0.7) +
scale_fill_manual(values = c("brown1", "mediumseagreen")) +
geom_dotplot(binaxis='y', stackdir='center', alpha=0.5, dotsize=0.5,
position=position_dodge(0.75))+
xlab("Plant:Tissue")+
ylab("%S")+
theme_classic() +
theme(axis.text.x = element_text(angle = 45, vjust = 1, hjust=1))
Zn.plot<- ggplot(plant.nut, aes(x=int.fac, y=ZN, fill=treatment)) +
geom_boxplot(alpha=0.7) +
scale_fill_manual(values = c("brown1", "mediumseagreen")) +
geom_dotplot(binaxis='y', stackdir='center', alpha=0.5, dotsize=0.5,
position=position_dodge(0.75)) +
xlab("Plant:Tissue")+
ylab("Zn ppm")+
theme_classic() +
theme(axis.text.x = element_text(angle = 45, vjust = 1, hjust=1))
extract.legend.nut <- get_legend(
# create some space to the left of the legend
Zn.plot + theme(legend.box.margin = margin(0, 0, 0, 10)))
nutrients<-plot_grid(
N.plot+ theme(legend.position = "none"),
P.plot+ theme(legend.position = "none"),
K.plot+ theme(legend.position = "none"),
S.plot+ theme(legend.position = "none"),
Zn.plot+ theme(legend.position = "none"),
extract.legend.nut,
rel_widths = c(8,8,8,8,8,3), ncol=6)
nutrients
ggsave("figures/leaf.nutrients.pdf", height=4, width=12)H2O Phosphorus
The phosphorous in water was only run at Time-2. Make a plot here and run the model.
phos<-read.csv("data/Pyro_water.phosph.csv")
fac2<-c("Time.Point", "Treatment", "Tank") # make factors
phos[fac2]<-lapply(phos[fac2],factor)
phos<-na.omit(phos)
############# phosphorous T2: significant smoothers, not treatment overall
m1.T2.phos<- gam(TP.umol..l ~ Treatment +
s(plant.mass..g, by=Treatment), data=phos, method="REML", family="gaussian")
m2.T2.phos<- gam(TP.umol..l ~ Treatment +
s(plant.mass..g), data=phos, method="REML", family="gaussian")
m3.T2.phos<- gam(TP.umol..l ~
s(plant.mass..g), data=phos, method="REML", family="gaussian")
AIC.T2.phos<-AIC(m1.T2.phos,m2.T2.phos, m3.T2.phos)
# factor smooth best
summary(m1.T2.phos)##
## Family: gaussian
## Link function: identity
##
## Formula:
## TP.umol..l ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 10.3649 0.5518 18.785 6.6e-12 ***
## Treatmentunburned -0.8831 0.7659 -1.153 0.267
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 2.924 3.525 154.7 <2e-16 ***
## s(plant.mass..g):Treatmentunburned 2.930 3.371 124.6 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## R-sq.(adj) = 0.979 Deviance explained = 98.5%
## -REML = 46.397 Scale est. = 2.9168 n = 23anova.gam(m1.T2.phos)##
## Family: gaussian
## Link function: identity
##
## Formula:
## TP.umol..l ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric Terms:
## df F p-value
## Treatment 1 1.329 0.267
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 2.924 3.525 154.7 <2e-16
## s(plant.mass..g):Treatmentunburned 2.930 3.371 124.6 <2e-16gam.check(m1.T2.phos, rep=1000)
##
## Method: REML Optimizer: outer newton
## full convergence after 4 iterations.
## Gradient range [-8.634731e-07,9.305738e-08]
## (score 46.39739 & scale 2.916807).
## Hessian positive definite, eigenvalue range [0.3315034,9.7044].
## Model rank = 20 / 20
##
## Basis dimension (k) checking results. Low p-value (k-index<1) may
## indicate that k is too low, especially if edf is close to k'.
##
## k' edf k-index p-value
## s(plant.mass..g):Treatmentburned 9.00 2.92 0.75 0.08 .
## s(plant.mass..g):Treatmentunburned 9.00 2.93 0.75 0.07 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1draw(m1.T2.phos)
concrvity(m1.T2.phos)## # A tibble: 9 × 3
## type term concurvity
## <chr> <chr> <dbl>
## 1 worst para 0.662
## 2 worst s(plant.mass..g):Treatmentburned 0.293
## 3 worst s(plant.mass..g):Treatmentunburned 1.00
## 4 observed para 0.662
## 5 observed s(plant.mass..g):Treatmentburned 0.0265
## 6 observed s(plant.mass..g):Treatmentunburned 0.0382
## 7 estimate para 0.662
## 8 estimate s(plant.mass..g):Treatmentburned 0.0410
## 9 estimate s(plant.mass..g):Treatmentunburned 0.0314par(mfrow = c(1, 2))
plot(m1.T2.phos, all.terms = TRUE, page=1)
# model predictions
phos.diff.T2<-plot_difference(
m1.T2.phos,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
phos.T2.mod.plot<-
plot_smooths(
model = m1.T2.phos,
series = plant.mass..g,
comparison = Treatment
) + theme(legend.position = "none") +
geom_point(data=phos,
aes(x=plant.mass..g, y=TP.umol..l, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
ylab(expression(paste("TP", ~(mu*mol/L), sep=""))) +
xlab("plant material (g)") +
ggtitle("Time-2") +
coord_cartesian(ylim=c(0, 50)) +
Fig.formatting +
theme(legend.key.size = unit(1,"line"))
phos.plots<-plot_grid(phos.T2.mod.plot, phos.diff.T2, ncol=2)
ggsave("figures/phos.plots.pdf", height=5, width=8)
## AIC table
mod.phos<-rep(c("Treatment + s(plant.mass..g, by=Treatment)",
"Treatment + s(plant.mass..g)",
"s(plant.mass..g)"), times=1)
mod.phos.df<- data.frame(mod.phos)
AIC.phos.mod<-cbind(mod.phos.df, AIC.T2.phos)
write.csv(AIC.phos.mod, "output/AIC models/AIC.phos.mod.csv")Greenhouse Gas Data
Import data and clean up
GHG<-read.csv("data/GH.gases/Pyro_ghg_headspace.csv")
GHG<-na.omit(GHG) # remove NAs
# set structure
make.fac<-c("Time.point", "Treatment", "Tank", "Gas")
GHG[make.fac] <- lapply(GHG[make.fac], factor) # make all these factors
GHG$plant.mass..g<-as.numeric(GHG$plant.mass..g)
# rename
GHG<- GHG %>% rename("GHG.nM" = "GHGwater..nmol.GHG..L.H2O")
# reorder columns and export
GHG<- GHG %>% select(Time.point, Treatment, plant.mass..g, Tank, Gas, GHG.nM)
GHG<-GHG[!(GHG$GHG.nM < 0),]Make plots and run models for CO2
CO2<-subset(GHG, Gas=="CO2")
#convert GHG units from nM to uM
CO2$GHG.uM<- CO2$GHG.nM/1000
############# GHG plots and modesl
#### CO2
#######-- T0
m1.T0.CO2<- gam(GHG.uM ~ Treatment +
s(plant.mass..g, by=Treatment), subset = Time.point=="T0", data = CO2,
method="REML", family="gaussian")
m2.T0.CO2<- gam(GHG.uM ~ Treatment +
s(plant.mass..g), subset = Time.point=="T0", data = CO2,
method="REML", family="gaussian")
m3.T0.CO2<- gam(GHG.uM ~
s(plant.mass..g), subset = Time.point=="T0", data = CO2,
method="REML", family="gaussian")
CO2.T0.AIC<-AIC(m1.T0.CO2,m2.T0.CO2, m3.T0.CO2)
#factor smooth best
summary(m1.T0.CO2)##
## Family: gaussian
## Link function: identity
##
## Formula:
## GHG.uM ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 34.4608 0.7814 44.099 < 2e-16 ***
## Treatmentunburned -4.2776 1.1051 -3.871 0.000776 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 1.000 1.00 10.566 0.00353 **
## s(plant.mass..g):Treatmentunburned 4.022 4.88 3.383 0.02381 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## R-sq.(adj) = 0.573 Deviance explained = 66.1%
## -REML = 74.819 Scale est. = 9.1598 n = 30anova.gam(m1.T0.CO2)##
## Family: gaussian
## Link function: identity
##
## Formula:
## GHG.uM ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric Terms:
## df F p-value
## Treatment 1 14.98 0.000776
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 1.000 1.000 10.566 0.00353
## s(plant.mass..g):Treatmentunburned 4.022 4.880 3.383 0.02381gam.check(m1.T0.CO2, rep=1000)
##
## Method: REML Optimizer: outer newton
## full convergence after 11 iterations.
## Gradient range [-1.834502e-05,7.70351e-07]
## (score 74.81935 & scale 9.159824).
## Hessian positive definite, eigenvalue range [1.834448e-05,13.18306].
## Model rank = 20 / 20
##
## Basis dimension (k) checking results. Low p-value (k-index<1) may
## indicate that k is too low, especially if edf is close to k'.
##
## k' edf k-index p-value
## s(plant.mass..g):Treatmentburned 9.00 1.00 1.03 0.48
## s(plant.mass..g):Treatmentunburned 9.00 4.02 1.03 0.47draw(m1.T0.CO2)
concrvity(m1.T0.CO2)## # A tibble: 9 × 3
## type term concurvity
## <chr> <chr> <dbl>
## 1 worst para 5.0 e- 1
## 2 worst s(plant.mass..g):Treatmentburned 1.98e-26
## 3 worst s(plant.mass..g):Treatmentunburned 2.09e-26
## 4 observed para 5.0 e- 1
## 5 observed s(plant.mass..g):Treatmentburned 9.97e-30
## 6 observed s(plant.mass..g):Treatmentunburned 3.55e-29
## 7 estimate para 5.00e- 1
## 8 estimate s(plant.mass..g):Treatmentburned 7.95e-29
## 9 estimate s(plant.mass..g):Treatmentunburned 7.49e-29par(mfrow = c(1, 2))
plot(m1.T0.CO2, all.terms = TRUE, page=1)
# model predictions
CO2.diff.T0<-plot_difference(
m1.T0.CO2,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
CO2.T0.mod.plot<-
plot_smooths(
model = m1.T0.CO2,
series = plant.mass..g,
comparison = Treatment
) + theme(legend.position = "none") +
geom_point(data=CO2[(CO2$Time.point=="T0"),],
aes(x=plant.mass..g, y=GHG.uM, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
ylab(expression(paste("CO"[2], ~(mu*M), sep=""))) +
xlab("plant material (g)") +
ggtitle("Time-0") +
coord_cartesian(ylim=c(0, 400)) +
Fig.formatting +
theme(legend.key.size = unit(1,"line"))
CO2.T0.mod.plot
#### CO2
#######-- T1
m1.T1.CO2<- gam(GHG.uM ~ Treatment +
s(plant.mass..g, by=Treatment), subset = Time.point=="T1", data = CO2,
method="REML", family="gaussian")
m2.T1.CO2<- gam(GHG.uM ~ Treatment +
s(plant.mass..g), subset = Time.point=="T1", data = CO2,
method="REML", family="gaussian")
m3.T1.CO2<- gam(GHG.uM ~
s(plant.mass..g), subset = Time.point=="T1", data = CO2,
method="REML", family="gaussian")
CO2.T1.AIC<-AIC(m1.T1.CO2,m2.T1.CO2, m3.T1.CO2)
#factor smooth best
summary(m1.T1.CO2)##
## Family: gaussian
## Link function: identity
##
## Formula:
## GHG.uM ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 137.112 4.830 28.387 < 2e-16 ***
## Treatmentunburned 20.946 6.831 3.066 0.00547 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 2.047 2.527 157.7 <2e-16 ***
## s(plant.mass..g):Treatmentunburned 2.966 3.633 144.2 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## R-sq.(adj) = 0.97 Deviance explained = 97.6%
## -REML = 121.27 Scale est. = 349.95 n = 30anova.gam(m1.T1.CO2)##
## Family: gaussian
## Link function: identity
##
## Formula:
## GHG.uM ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric Terms:
## df F p-value
## Treatment 1 9.403 0.00547
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 2.047 2.527 157.7 <2e-16
## s(plant.mass..g):Treatmentunburned 2.966 3.633 144.2 <2e-16gam.check(m1.T1.CO2, rep=1000)
##
## Method: REML Optimizer: outer newton
## full convergence after 4 iterations.
## Gradient range [-1.86084e-07,1.75551e-07]
## (score 121.2725 & scale 349.9528).
## Hessian positive definite, eigenvalue range [0.1792501,13.09977].
## Model rank = 20 / 20
##
## Basis dimension (k) checking results. Low p-value (k-index<1) may
## indicate that k is too low, especially if edf is close to k'.
##
## k' edf k-index p-value
## s(plant.mass..g):Treatmentburned 9.00 2.05 1.02 0.52
## s(plant.mass..g):Treatmentunburned 9.00 2.97 1.02 0.55draw(m1.T1.CO2)
concrvity(m1.T1.CO2)## # A tibble: 9 × 3
## type term concurvity
## <chr> <chr> <dbl>
## 1 worst para 5.0 e- 1
## 2 worst s(plant.mass..g):Treatmentburned 1.98e-26
## 3 worst s(plant.mass..g):Treatmentunburned 2.09e-26
## 4 observed para 5.00e- 1
## 5 observed s(plant.mass..g):Treatmentburned 1.01e-29
## 6 observed s(plant.mass..g):Treatmentunburned 2.32e-30
## 7 estimate para 5.00e- 1
## 8 estimate s(plant.mass..g):Treatmentburned 7.95e-29
## 9 estimate s(plant.mass..g):Treatmentunburned 7.49e-29par(mfrow = c(1, 2))
plot(m1.T1.CO2, all.terms = TRUE, page=1)
# model predictions
CO2.diff.T1<-plot_difference(
m1.T1.CO2,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
CO2.T1.mod.plot<-
plot_smooths(
model = m1.T1.CO2,
series = plant.mass..g,
comparison = Treatment
) + theme(legend.position = "none") +
geom_point(data=CO2[(CO2$Time.point=="T1"),],
aes(x=plant.mass..g, y=GHG.uM, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
ylab(expression(paste("CO"[2], ~(mu*M), sep=""))) +
xlab("plant material (g)") +
ggtitle("Time-1") +
coord_cartesian(ylim=c(0, 400)) +
Fig.formatting +
theme(legend.key.size = unit(1,"line"))
CO2.T1.mod.plot
#### CO2
#######-- T2
m1.T2.CO2<- gam(GHG.uM ~ Treatment +
s(plant.mass..g, by=Treatment), subset = Time.point=="T2", data = CO2,
method="REML", family="gaussian")
m2.T2.CO2<- gam(GHG.uM ~ Treatment +
s(plant.mass..g), subset = Time.point=="T2", data = CO2,
method="REML", family="gaussian")
m3.T2.CO2<- gam(GHG.uM ~
s(plant.mass..g), subset = Time.point=="T2", data = CO2,
method="REML", family="gaussian")
CO2.T2.AIC<-AIC(m1.T2.CO2, m2.T2.CO2, m3.T2.CO2)
# smooth by factor best
summary(m1.T2.CO2)##
## Family: gaussian
## Link function: identity
##
## Formula:
## GHG.uM ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 69.394 4.628 14.995 4.65e-13 ***
## Treatmentunburned 4.275 6.545 0.653 0.52
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 2.499 3.072 52.20 <2e-16 ***
## s(plant.mass..g):Treatmentunburned 3.422 4.176 22.47 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## R-sq.(adj) = 0.897 Deviance explained = 92.2%
## -REML = 121.41 Scale est. = 321.23 n = 30anova.gam(m1.T2.CO2)##
## Family: gaussian
## Link function: identity
##
## Formula:
## GHG.uM ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric Terms:
## df F p-value
## Treatment 1 0.427 0.52
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 2.499 3.072 52.20 <2e-16
## s(plant.mass..g):Treatmentunburned 3.422 4.176 22.47 <2e-16gam.check(m1.T2.CO2, rep=1000)
##
## Method: REML Optimizer: outer newton
## full convergence after 3 iterations.
## Gradient range [-9.205725e-07,1.593428e-06]
## (score 121.4081 & scale 321.2311).
## Hessian positive definite, eigenvalue range [0.3528807,13.16313].
## Model rank = 20 / 20
##
## Basis dimension (k) checking results. Low p-value (k-index<1) may
## indicate that k is too low, especially if edf is close to k'.
##
## k' edf k-index p-value
## s(plant.mass..g):Treatmentburned 9.00 2.50 0.93 0.26
## s(plant.mass..g):Treatmentunburned 9.00 3.42 0.93 0.33draw(m1.T2.CO2)
concrvity(m1.T2.CO2)## # A tibble: 9 × 3
## type term concurvity
## <chr> <chr> <dbl>
## 1 worst para 5.0 e- 1
## 2 worst s(plant.mass..g):Treatmentburned 1.98e-26
## 3 worst s(plant.mass..g):Treatmentunburned 2.09e-26
## 4 observed para 5.0 e- 1
## 5 observed s(plant.mass..g):Treatmentburned 8.86e-30
## 6 observed s(plant.mass..g):Treatmentunburned 3.18e-30
## 7 estimate para 5.00e- 1
## 8 estimate s(plant.mass..g):Treatmentburned 7.95e-29
## 9 estimate s(plant.mass..g):Treatmentunburned 7.49e-29par(mfrow = c(1, 2))
plot(m1.T2.CO2, all.terms = TRUE, page=1)
# model predictions
CO2.diff.T2<-plot_difference(
m1.T2.CO2,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
CO2.T2.mod.plot<-
plot_smooths(
model = m1.T2.CO2,
series = plant.mass..g,
comparison = Treatment
) + theme(legend.position = "none") +
geom_point(data=CO2[(CO2$Time.point=="T2"),],
aes(x=plant.mass..g, y=GHG.uM, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
ylab(expression(paste("CO"[2], ~(mu*M), sep=""))) +
xlab("plant material (g)") +
ggtitle("Time-2") +
coord_cartesian(ylim=c(0, 400)) +
Fig.formatting +
theme(legend.key.size = unit(1,"line"))
CO2.T2.mod.plot
#### CO2
#######-- T3
m1.T3.CO2<- gam(GHG.uM ~ Treatment +
s(plant.mass..g, by=Treatment), subset = Time.point=="T3", data = CO2,
method="REML", family="gaussian")
m2.T3.CO2<- gam(GHG.uM ~ Treatment +
s(plant.mass..g), subset = Time.point=="T3", data = CO2,
method="REML", family="gaussian")
m3.T3.CO2<- gam(GHG.uM ~
s(plant.mass..g), subset = Time.point=="T3", data = CO2,
method="REML", family="gaussian")
CO2.T3.AIC<-AIC(m1.T3.CO2, m2.T3.CO2, m3.T3.CO2)
# smooth by factor best
summary(m1.T3.CO2)##
## Family: gaussian
## Link function: identity
##
## Formula:
## GHG.uM ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 36.020 5.742 6.273 2.02e-06 ***
## Treatmentunburned 12.209 7.980 1.530 0.14
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 2.744 3.366 11.86 4.97e-05 ***
## s(plant.mass..g):Treatmentunburned 1.000 1.000 28.09 2.26e-05 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## R-sq.(adj) = 0.705 Deviance explained = 75.5%
## -REML = 119.18 Scale est. = 460.52 n = 29anova.gam(m1.T3.CO2)##
## Family: gaussian
## Link function: identity
##
## Formula:
## GHG.uM ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric Terms:
## df F p-value
## Treatment 1 2.341 0.14
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 2.744 3.366 11.86 4.97e-05
## s(plant.mass..g):Treatmentunburned 1.000 1.000 28.09 2.26e-05gam.check(m1.T3.CO2, rep=1000)
##
## Method: REML Optimizer: outer newton
## full convergence after 9 iterations.
## Gradient range [-3.53958e-05,8.858783e-05]
## (score 119.1836 & scale 460.5193).
## Hessian positive definite, eigenvalue range [3.540667e-05,12.56498].
## Model rank = 20 / 20
##
## Basis dimension (k) checking results. Low p-value (k-index<1) may
## indicate that k is too low, especially if edf is close to k'.
##
## k' edf k-index p-value
## s(plant.mass..g):Treatmentburned 9.00 2.74 0.96 0.32
## s(plant.mass..g):Treatmentunburned 9.00 1.00 0.96 0.40draw(m1.T3.CO2)
concrvity(m1.T3.CO2)## # A tibble: 9 × 3
## type term concurvity
## <chr> <chr> <dbl>
## 1 worst para 0.536
## 2 worst s(plant.mass..g):Treatmentburned 0.0380
## 3 worst s(plant.mass..g):Treatmentunburned 0.0113
## 4 observed para 0.536
## 5 observed s(plant.mass..g):Treatmentburned 0.000735
## 6 observed s(plant.mass..g):Treatmentunburned 0.000138
## 7 estimate para 0.536
## 8 estimate s(plant.mass..g):Treatmentburned 0.000272
## 9 estimate s(plant.mass..g):Treatmentunburned 0.000250par(mfrow = c(1, 2))
plot(m1.T3.CO2, all.terms = TRUE, page=1)
# model predictions
CO2.diff.T3<-plot_difference(
m1.T3.CO2,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
CO2.T3.mod.plot<-
plot_smooths(
model = m1.T3.CO2,
series = plant.mass..g,
comparison = Treatment
) + theme(legend.position = "none") +
geom_point(data=CO2[(CO2$Time.point=="T3"),],
aes(x=plant.mass..g, y=GHG.uM, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
ylab(expression(paste("CO"[2], ~(mu*M), sep=""))) +
xlab("plant material (g)") +
ggtitle("Time-3") +
coord_cartesian(ylim=c(0, 400)) +
Fig.formatting +
theme(legend.key.size = unit(1,"line"))
CO2.T3.mod.plot
# models and raw data
CO2.model.plots<-plot_grid(
CO2.T0.mod.plot+ theme(legend.position= "none") + ggtitle( "Day-0"),
CO2.T1.mod.plot+ theme(legend.position= "none") + ggtitle( "Day-10"),
CO2.T2.mod.plot+ theme(legend.position= "none") + ggtitle( "Day-31"),
CO2.T3.mod.plot+ theme(legend.position= "none") + ggtitle( "Day-59"),
extract.legend,
rel_widths = c(8,8,8,8,3), ncol=5)
### model differences
CO2.plot.diff<-plot_grid(
CO2.diff.T0+ ggtitle("CO2-T0"),
CO2.diff.T1+ ggtitle("Day-10"),
CO2.diff.T2+ ggtitle("Day-31"),
CO2.diff.T3+ ggtitle("Day-59"),
rel_widths = c(8,8,8,8), ncol=4)
## AIC table
mod.ghg<-rep(c("Treatment +s(plant.mass..g, by=Treatment)",
"Treatment + s(plant.mass..g)",
"s(plant.mass..g)"), times=4)
mod.ghg.df<- data.frame(mod.ghg)
AIC.CO2<-bind_rows(CO2.T0.AIC, CO2.T1.AIC, CO2.T2.AIC, CO2.T3.AIC)
AIC.CO2.mod<-cbind(mod.ghg.df, AIC.CO2)
write.csv(AIC.CO2.mod, "output/AIC models/AIC.CO2.mod.csv")Make plots and run models for methane.
CH4<-subset(GHG, Gas=="CH4")
m1.T0.CH4<- gam(GHG.nM ~ Treatment +
s(plant.mass..g, by=Treatment), subset = Time.point=="T0", data = CH4,
method="REML", family="gaussian")
m2.T0.CH4<- gam(GHG.nM ~ Treatment +
s(plant.mass..g), subset = Time.point=="T0", data = CH4,
method="REML", family="gaussian")
m3.T0.CH4<- gam(GHG.nM ~
s(plant.mass..g), subset = Time.point=="T0", data = CH4,
method="REML", family="gaussian")
CH4.T0.AIC<-AIC(m1.T0.CH4, m2.T0.CH4, m3.T0.CH4)
#factor single smooth best
summary(m2.T0.CH4)##
## Family: gaussian
## Link function: identity
##
## Formula:
## GHG.nM ~ Treatment + s(plant.mass..g)
##
## Parametric coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 4.3461 0.6747 6.441 1.16e-06 ***
## Treatmentunburned 1.3989 0.9799 1.427 0.166
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g) 1 1 0.718 0.405
##
## R-sq.(adj) = 0.0155 Deviance explained = 9.13%
## -REML = 60.338 Scale est. = 6.2802 n = 27anova.gam(m2.T0.CH4)##
## Family: gaussian
## Link function: identity
##
## Formula:
## GHG.nM ~ Treatment + s(plant.mass..g)
##
## Parametric Terms:
## df F p-value
## Treatment 1 2.038 0.166
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g) 1 1 0.718 0.405gam.check(m2.T0.CH4, rep=1000)
##
## Method: REML Optimizer: outer newton
## full convergence after 6 iterations.
## Gradient range [-1.631119e-05,8.871995e-05]
## (score 60.33814 & scale 6.280154).
## Hessian positive definite, eigenvalue range [1.632563e-05,11.99991].
## Model rank = 11 / 11
##
## Basis dimension (k) checking results. Low p-value (k-index<1) may
## indicate that k is too low, especially if edf is close to k'.
##
## k' edf k-index p-value
## s(plant.mass..g) 9 1 1.13 0.68draw(m2.T0.CH4)
concrvity(m2.T0.CH4)## # A tibble: 6 × 3
## type term concurvity
## <chr> <chr> <dbl>
## 1 worst para 0.499
## 2 worst s(plant.mass..g) 0.0659
## 3 observed para 0.499
## 4 observed s(plant.mass..g) 0.0298
## 5 estimate para 0.499
## 6 estimate s(plant.mass..g) 0.0232par(mfrow = c(1, 2))
plot(m2.T0.CH4, all.terms = TRUE, page=1)
# model predictions
CH4.diff.T0<-plot_difference(
m2.T0.CH4,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
CH4.T0.mod.plot<-
plot_smooths(
model = m2.T0.CH4,
series = plant.mass..g,
comparison=Treatment
) + theme(legend.position = "none") +
geom_point(data=CH4[(CH4$Time.point=="T0"),],
aes(x=plant.mass..g, y=GHG.nM, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
geom_line(aes(fill=Treatment, linetype=Treatment)) +
ylab(expression(paste("CH"[4], ~(nM), sep=""))) +
xlab("plant material (g)") +
ggtitle("Time-0") +
coord_cartesian(ylim=c(0, 50)) +
Fig.formatting +
theme(legend.key.size = unit(1,"line"))
CH4.T0.mod.plot
#### CH4
#######-- T1
m1.T1.CH4<- gam(GHG.nM ~ Treatment +
s(plant.mass..g, by=Treatment), subset = Time.point=="T1", data = CH4,
method="REML", family="gaussian")
m2.T1.CH4<- gam(GHG.nM ~ Treatment +
s(plant.mass..g), subset = Time.point=="T1", data = CH4,
method="REML", family="gaussian")
m3.T1.CH4<- gam(GHG.nM ~
s(plant.mass..g), subset = Time.point=="T1", data = CH4,
method="REML", family="gaussian")
CH4.T1.AIC<-AIC(m1.T1.CH4, m2.T1.CH4, m3.T1.CH4)
#factor by smooth best
summary(m1.T1.CH4)##
## Family: gaussian
## Link function: identity
##
## Formula:
## GHG.nM ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 11.6731 0.7459 15.649 1.07e-13 ***
## Treatmentunburned 0.5438 1.0549 0.516 0.611
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 1.813 2.244 0.89 0.42743
## s(plant.mass..g):Treatmentunburned 3.346 4.086 6.53 0.00119 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## R-sq.(adj) = 0.476 Deviance explained = 58.7%
## -REML = 73.049 Scale est. = 8.3465 n = 30anova.gam(m1.T1.CH4)##
## Family: gaussian
## Link function: identity
##
## Formula:
## GHG.nM ~ Treatment + s(plant.mass..g, by = Treatment)
##
## Parametric Terms:
## df F p-value
## Treatment 1 0.266 0.611
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g):Treatmentburned 1.813 2.244 0.89 0.42743
## s(plant.mass..g):Treatmentunburned 3.346 4.086 6.53 0.00119gam.check(m1.T1.CH4, rep=1000)
##
## Method: REML Optimizer: outer newton
## full convergence after 5 iterations.
## Gradient range [-2.015212e-07,2.357495e-08]
## (score 73.04861 & scale 8.346463).
## Hessian positive definite, eigenvalue range [0.2435469,13.121].
## Model rank = 20 / 20
##
## Basis dimension (k) checking results. Low p-value (k-index<1) may
## indicate that k is too low, especially if edf is close to k'.
##
## k' edf k-index p-value
## s(plant.mass..g):Treatmentburned 9.00 1.81 1.04 0.56
## s(plant.mass..g):Treatmentunburned 9.00 3.35 1.04 0.56draw(m1.T1.CH4)
concrvity(m1.T1.CH4)## # A tibble: 9 × 3
## type term concurvity
## <chr> <chr> <dbl>
## 1 worst para 5.0 e- 1
## 2 worst s(plant.mass..g):Treatmentburned 1.98e-26
## 3 worst s(plant.mass..g):Treatmentunburned 2.09e-26
## 4 observed para 5.00e- 1
## 5 observed s(plant.mass..g):Treatmentburned 1.11e-30
## 6 observed s(plant.mass..g):Treatmentunburned 4.58e-30
## 7 estimate para 5.00e- 1
## 8 estimate s(plant.mass..g):Treatmentburned 7.95e-29
## 9 estimate s(plant.mass..g):Treatmentunburned 7.49e-29par(mfrow = c(1, 2))
plot(m1.T1.CH4, all.terms = TRUE, page=1)
# model predictions
CH4.diff.T1<-plot_difference(
m1.T1.CH4,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
CH4.T1.mod.plot<-
plot_smooths(
model = m1.T1.CH4,
series = plant.mass..g,
comparison = Treatment
) + theme(legend.position = "none") +
geom_point(data=CH4[(CH4$Time.point=="T1"),],
aes(x=plant.mass..g, y=GHG.nM, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
ylab(expression(paste("CH"[4], ~(mu*M), sep=""))) +
xlab("plant material (g)") +
ggtitle("Time-1") +
coord_cartesian(ylim=c(0, 50)) +
Fig.formatting +
theme(legend.key.size = unit(1,"line"))
CH4.T1.mod.plot
#### CH4
#######-- T2
m1.T2.CH4<- gam(GHG.nM ~ Treatment +
s(plant.mass..g, by=Treatment), subset = Time.point=="T2", data = CH4,
method="REML", family="gaussian")
m2.T2.CH4<- gam(GHG.nM ~ Treatment +
s(plant.mass..g), subset = Time.point=="T2", data = CH4,
method="REML", family="gaussian")
m3.T2.CH4<- gam(GHG.nM ~
s(plant.mass..g), subset = Time.point=="T2", data = CH4,
method="REML", family="gaussian")
CH4.T2.AIC<-AIC(m1.T2.CH4, m2.T2.CH4, m3.T2.CH4)
# global best
summary(m3.T2.CH4)##
## Family: gaussian
## Link function: identity
##
## Formula:
## GHG.nM ~ s(plant.mass..g)
##
## Parametric coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 9.628 1.066 9.031 8.69e-10 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g) 1 1.001 0.19 0.667
##
## R-sq.(adj) = -0.0287 Deviance explained = 0.675%
## -REML = 92.541 Scale est. = 34.099 n = 30anova.gam(m3.T2.CH4)##
## Family: gaussian
## Link function: identity
##
## Formula:
## GHG.nM ~ s(plant.mass..g)
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g) 1.000 1.001 0.19 0.667gam.check(m3.T2.CH4, rep=1000)
##
## Method: REML Optimizer: outer newton
## full convergence after 11 iterations.
## Gradient range [-2.147014e-05,0.000126711]
## (score 92.54116 & scale 34.09853).
## Hessian positive definite, eigenvalue range [2.148928e-05,13.99987].
## Model rank = 10 / 10
##
## Basis dimension (k) checking results. Low p-value (k-index<1) may
## indicate that k is too low, especially if edf is close to k'.
##
## k' edf k-index p-value
## s(plant.mass..g) 9 1 0.84 0.16draw(m3.T2.CH4)
concrvity(m3.T2.CH4)## # A tibble: 6 × 3
## type term concurvity
## <chr> <chr> <dbl>
## 1 worst para 2.08e-26
## 2 worst s(plant.mass..g) 2.09e-26
## 3 observed para 2.08e-26
## 4 observed s(plant.mass..g) 2.62e-32
## 5 estimate para 2.08e-26
## 6 estimate s(plant.mass..g) 7.15e-29par(mfrow = c(1, 2))
plot(m3.T2.CH4, all.terms = TRUE, page=1)
# model predictions
CH4.diff.T2<-plot_difference(
m1.T2.CH4,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
CH4.T2.mod.plot<-
plot_smooths(
model = m3.T2.CH4,
series = plant.mass..g,
) + theme(legend.position = "none") +
geom_point(data=CH4[(CH4$Time.point=="T2"),],
aes(x=plant.mass..g, y=GHG.nM, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
ylab(expression(paste("CH"[4], ~(mu*M), sep=""))) +
xlab("plant material (g)") +
ggtitle("Time-2") +
coord_cartesian(ylim=c(0, 50)) +
Fig.formatting +
theme(legend.key.size = unit(1,"line"))
CH4.T2.mod.plot
#### CH4
#######-- T3
m1.T3.CH4<- gam(GHG.nM ~ Treatment +
s(plant.mass..g, by=Treatment), subset = Time.point=="T3", data = CH4,
method="REML", family="gaussian")
m2.T3.CH4<- gam(GHG.nM ~ Treatment +
s(plant.mass..g), subset = Time.point=="T3", data = CH4,
method="REML", family="gaussian")
m3.T3.CH4<- gam(GHG.nM ~
s(plant.mass..g), subset = Time.point=="T3", data = CH4,
method="REML", family="gaussian")
CH4.T3.AIC<-AIC(m1.T3.CH4, m2.T3.CH4, m3.T3.CH4)
# global with treatment best
summary(m2.T3.CH4)##
## Family: gaussian
## Link function: identity
##
## Formula:
## GHG.nM ~ Treatment + s(plant.mass..g)
##
## Parametric coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 22.395 2.255 9.930 4.33e-10 ***
## Treatmentunburned -6.089 3.189 -1.909 0.068 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g) 3.381 4.127 2.038 0.113
##
## R-sq.(adj) = 0.266 Deviance explained = 37.7%
## -REML = 103.88 Scale est. = 76.293 n = 30anova.gam(m2.T3.CH4)##
## Family: gaussian
## Link function: identity
##
## Formula:
## GHG.nM ~ Treatment + s(plant.mass..g)
##
## Parametric Terms:
## df F p-value
## Treatment 1 3.645 0.068
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(plant.mass..g) 3.381 4.127 2.038 0.113gam.check(m2.T3.CH4, rep=1000)
##
## Method: REML Optimizer: outer newton
## full convergence after 6 iterations.
## Gradient range [-1.041682e-09,9.185275e-11]
## (score 103.8784 & scale 76.2932).
## Hessian positive definite, eigenvalue range [0.4424777,13.60852].
## Model rank = 11 / 11
##
## Basis dimension (k) checking results. Low p-value (k-index<1) may
## indicate that k is too low, especially if edf is close to k'.
##
## k' edf k-index p-value
## s(plant.mass..g) 9.00 3.38 1.15 0.8draw(m2.T3.CH4)
concrvity(m2.T3.CH4)## # A tibble: 6 × 3
## type term concurvity
## <chr> <chr> <dbl>
## 1 worst para 5 e- 1
## 2 worst s(plant.mass..g) 2.09e-26
## 3 observed para 5 e- 1
## 4 observed s(plant.mass..g) 5.75e-31
## 5 estimate para 5.0 e- 1
## 6 estimate s(plant.mass..g) 7.15e-29par(mfrow = c(1, 2))
plot(m2.T3.CH4, all.terms = TRUE, page=1)
# model predictions
CH4.diff.T3<-plot_difference(
m1.T3.CH4,
series = plant.mass..g,
difference = list(Treatment = c("burned", "unburned"))
)
###########
#plot for the model output on rawdata
CH4.T3.mod.plot<-
plot_smooths(
model = m2.T3.CH4,
series = plant.mass..g,
comparison=Treatment
) + theme(legend.position = "none") +
geom_point(data=CH4[(CH4$Time.point=="T3"),],
aes(x=plant.mass..g, y=GHG.nM, color=Treatment)) +
scale_color_manual(values = c("brown1", "mediumseagreen")) +
geom_line(aes(fill=Treatment, linetype=Treatment)) +
ylab(expression(paste("CH"[4], ~(nM), sep=""))) +
xlab("plant material (g)") +
ggtitle("Time-3") +
coord_cartesian(ylim=c(0, 50)) +
Fig.formatting +
theme(legend.key.size = unit(1,"line"))
CH4.model.plots<-plot_grid(
CH4.T0.mod.plot+ theme(legend.position= "none") + ggtitle( "Day-0"),
CH4.T1.mod.plot+ theme(legend.position= "none") + ggtitle( "Day-10"),
CH4.T2.mod.plot+ theme(legend.position= "none") + ggtitle( "Day-31"),
CH4.T3.mod.plot+ theme(legend.position= "none") + ggtitle( "Day-59"),
extract.legend,
rel_widths = c(8,8,8,8,3), ncol=5)
### model differences
CH4.plot.diff<-plot_grid(
CH4.diff.T0+ ggtitle("CH4-Day-0"),
CH4.diff.T1+ ggtitle("Day-10"),
CH4.diff.T2+ ggtitle("Day-31"),
CH4.diff.T3+ ggtitle("Day-59"),
rel_widths = c(8,8,8,8), ncol=4)
##### AIC table
AIC.CH4<-bind_rows(CH4.T0.AIC, CH4.T1.AIC, CH4.T2.AIC, CH4.T3.AIC)
AIC.CH4.mod<-cbind(mod.ghg.df, AIC.CH4)
write.csv(AIC.CH4.mod, "output/AIC models/AIC.CH4.mod.csv")Compile the greenhouse gas plots
GHG.plots<-plot_grid(CO2.model.plots, CH4.model.plots, ncol=1)
ggsave("figures/GHG.molar.mod.plots.pdf", height=7, width=12)
GHG.plot.diff<-plot_grid(CO2.plot.diff, CH4.plot.diff, ncol=1)
ggsave("figures/GHG.molar.mod.diff.pdf", height=7, width=12)